Загрузка видео...

Не удалось загрузить видео

На главную

Presenting Unsupervised Actuator Nets (UANs) that push the limits of agile whole-body control without the need for reward shaping! ⚡️ UANs reduce the sim2real gap in robot's motors removing the need for reward design to bridge the sim2real gap. ⚡️ A pre-trained whole-body controller uses reference motion as a...

pulkitology's profile picture

Pulkit Agrawal

16,675 subscribers

11,196 просмотров • 1 год назад •via X (Twitter)

Наука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 3

Фото профиля Pulkit Agrawal
Pulkit Agrawal1 год назад

Pushing the limits requires brave souls!

Фото профиля NICE
NICE1 год назад

Stay competitive by balancing cutting-edge AI with automation tools. Forrester shows how.

Фото профиля Ray Kuo
Ray Kuo1 год назад

cool!

Похожие видео

TWIST: a real-time teleoperation system for humanoid robots to mimic whole-body motions. Reference motion data is generated by retargeting human motion-capture data to the robot. Then, the controller is trained in simulation using reinforcement learning and behavior cloning.
0:58

TWIST: a real-time teleoperation system for humanoid robots to mimic whole-body motions. Reference motion data is generated by retargeting human motion-capture data to the robot. Then, the controller is trained in simulation using reinforcement learning and behavior cloning.

The Humanoid Hub

46,715 просмотров • 1 год назад

This work is not about a new technique. GMT (General Motion Tracking) shows good engineering practices that you can actually train a single unified whole-body control policy for all agile motion, and it works in the real world, directly with sim2real without adaptation. This is different from many existing works that trains/adapts one individual policy for each specific motion. Instead of just showing the beautiful motions/dances, this is actually practical. Because now you can convert all types of commands into real actions in real time without tuning anything. Visit to check out how we handle the following challenges: ✅ Partial Observability. ✅ Hardware Limitations. ✅ Unbalanced Data Distribution. ✅ Model Expressiveness. Work led by Zixuan Chen Mazeyu Ji
1:28

This work is not about a new technique. GMT (General Motion Tracking) shows good engineering practices that you can actually train a single unified whole-body control policy for all agile motion, and it works in the real world, directly with sim2real without adaptation. This is different from many existing works that trains/adapts one individual policy for each specific motion. Instead of just showing the beautiful motions/dances, this is actually practical. Because now you can convert all types of commands into real actions in real time without tuning anything. Visit to check out how we handle the following challenges: ✅ Partial Observability. ✅ Hardware Limitations. ✅ Unbalanced Data Distribution. ✅ Model Expressiveness. Work led by Zixuan Chen Mazeyu Ji

Xiaolong Wang

83,347 просмотров • 1 год назад

I've been playing more with this depth map motion transfer workflow, and if you're trying to recreate really specific body movements from a reference video, this works surprisingly well. Here's the workflow: 1. Grab your reference video, then create your character and environment references. 2. Run the video through Depth Anything on fal to generate a depth map: 3. Drop that depth map into Seedance Omni as your motion reference along with your character and environment references. You don't even need the original video. I also tested this against using the original video as the motion reference in Seedance, and there's definitely a noticeable difference. One thing I found, though, is that this works much better for body movement than facial performance since the depth map can't really capture subtle facial motion. I'm running more tests to see if there's a good way to control both body movement and facial performance. If you have any workflows I should try, do share in the comments!
0:52

I've been playing more with this depth map motion transfer workflow, and if you're trying to recreate really specific body movements from a reference video, this works surprisingly well. Here's the workflow: 1. Grab your reference video, then create your character and environment references. 2. Run the video through Depth Anything on fal to generate a depth map: 3. Drop that depth map into Seedance Omni as your motion reference along with your character and environment references. You don't even need the original video. I also tested this against using the original video as the motion reference in Seedance, and there's definitely a noticeable difference. One thing I found, though, is that this works much better for body movement than facial performance since the depth map can't really capture subtle facial motion. I'm running more tests to see if there's a good way to control both body movement and facial performance. If you have any workflows I should try, do share in the comments!

Curious Refuge

16,487 просмотров • 1 месяц назад

We are excited to share our #CORL2024 paper on learning quadrotor obstacle avoidance from the visual stream of a single #eventcamera! Trained entirely in simulation! We demonstrate obstacle avoidance both in the dark and in a forest up to 5m/s. PDF: Video: Project page: Event cameras are sensors that output per-pixel-level intensity changes at microsecond latency resolution; they feature nearly zero motion blur and high dynamic range but produce a very large volume of events under significant ego-motion and further lack a high-fidelity continuous-time sensor model in simulation, making direct #sim2real transfer not possible. By leveraging depth prediction as a pretext task, we pre-train a reactive obstacle avoidance policy with “approximated” simulated events and then fine-tune the perception component with limited events-and-depth real-world data. This technique bridges the sim2real gap for #eventcameras! As at the current state, there is no continuous-time sensor model for event cameras, we hope that this work can finally spur future research leveraging simulation for training event-vision-based policies to create faster, agile robots! Kudos to Anish Bhattacharya, @marcocannic, Vijay Kumar Nikolai Matni UZH Science University of Zurich UZH Space Hub UZH IfI European Research Council (ERC) GRASP Laboratory Penn Engineering
1:25

We are excited to share our #CORL2024 paper on learning quadrotor obstacle avoidance from the visual stream of a single #eventcamera! Trained entirely in simulation! We demonstrate obstacle avoidance both in the dark and in a forest up to 5m/s. PDF: Video: Project page: Event cameras are sensors that output per-pixel-level intensity changes at microsecond latency resolution; they feature nearly zero motion blur and high dynamic range but produce a very large volume of events under significant ego-motion and further lack a high-fidelity continuous-time sensor model in simulation, making direct #sim2real transfer not possible. By leveraging depth prediction as a pretext task, we pre-train a reactive obstacle avoidance policy with “approximated” simulated events and then fine-tune the perception component with limited events-and-depth real-world data. This technique bridges the sim2real gap for #eventcameras! As at the current state, there is no continuous-time sensor model for event cameras, we hope that this work can finally spur future research leveraging simulation for training event-vision-based policies to create faster, agile robots! Kudos to Anish Bhattacharya, @marcocannic, Vijay Kumar Nikolai Matni UZH Science University of Zurich UZH Space Hub UZH IfI European Research Council (ERC) GRASP Laboratory Penn Engineering

Davide Scaramuzza

17,219 просмотров • 1 год назад

I have been cleaning my daughter's mess for more than two years now. Last weekend our robot came to home to do the job for me. 🤖 Our new work on visual whole-body control learns a policy to coordinate the robot legs and arms for mobile manipulation. See how the legs bend on grasping objects on the ground based on visual inputs autonomously. (It is NOT teleoperation!) We again adopt a Sim2Real approach and our policy generalizes to beaches, forests, and streets in San Diego. 👇🧵
3:15

I have been cleaning my daughter's mess for more than two years now. Last weekend our robot came to home to do the job for me. 🤖 Our new work on visual whole-body control learns a policy to coordinate the robot legs and arms for mobile manipulation. See how the legs bend on grasping objects on the ground based on visual inputs autonomously. (It is NOT teleoperation!) We again adopt a Sim2Real approach and our policy generalizes to beaches, forests, and streets in San Diego. 👇🧵

Xiaolong Wang

159,462 просмотров • 2 лет назад

IF YOU MISSED BEING EARLY IN CONTROL TOKEN WE ARE PROVIDING THAT OPPORTUNITY AGAIN! Our New Revolutionary Reward System Provides the ability to become a whale using the same token without competing with the previous holders. Only NEW Buyers can build a Position in this New Tier of Rewards Using the Same Token while Supporting the growth of the Project as a whole. More Coming Soon on this. CA:0x38c369055918797144cab3305f87C7cf7b6Be1A8
5:28

IF YOU MISSED BEING EARLY IN CONTROL TOKEN WE ARE PROVIDING THAT OPPORTUNITY AGAIN! Our New Revolutionary Reward System Provides the ability to become a whale using the same token without competing with the previous holders. Only NEW Buyers can build a Position in this New Tier of Rewards Using the Same Token while Supporting the growth of the Project as a whole. More Coming Soon on this. CA:0x38c369055918797144cab3305f87C7cf7b6Be1A8

Crypto Control

19,276 просмотров • 1 год назад

Big news from the TeleAI embodied intelligence team! They unveiled TextOp, a general cerebellar framework for real-time, text-driven humanoid control. TextOp lets humans control the robot via natural language, dynamically modifying instructions during runtime to instantly generate smooth, whole-body actions. This enables precise and highly versatile control. The framework uses a two-layer architecture: an action diffusion model (High-Level) and a general motion tracking policy (Low-Level). It’s a new paradigm that eliminates the need for pre-recorded scripts or manual programming.
1:20

Big news from the TeleAI embodied intelligence team! They unveiled TextOp, a general cerebellar framework for real-time, text-driven humanoid control. TextOp lets humans control the robot via natural language, dynamically modifying instructions during runtime to instantly generate smooth, whole-body actions. This enables precise and highly versatile control. The framework uses a two-layer architecture: an action diffusion model (High-Level) and a general motion tracking policy (Low-Level). It’s a new paradigm that eliminates the need for pre-recorded scripts or manual programming.

RoboHub🤖

24,297 просмотров • 10 месяцев назад

We RL'ed humanoid robots to Cristiano Ronaldo, LeBron James, and Kobe Byrant! These are neural nets running on real hardware at our GEAR lab. Most robot demos you see online speed videos up. We actually *slow them down* so you can enjoy the fluid motions. I'm excited to announce "ASAP", a "real2sim2real" model that masters extremely smooth and dynamic motions for humanoid whole body control. We pretrain the robot in simulation first, but there is a notorious "sim2real" gap: it's very difficult for hand-engineered physics equations to match real world dynamics. Our fix is simple: just deploy a pretrained policy on real hardware, collect data, and replay the motion in sim. The replay will obviously have many errors, but that gives a rich signal to compensate for the physics discrepancy. Use another neural net to learn the delta. Basically, we "patch up" a traditional physics engine, so that the robot can experience almost the real world at scale in GPUs. The future is hybrid simulation: combine the power of classical sim engines refined over decades and the uncanny ability of modern NNs to capture a messy world.
0:50

We RL'ed humanoid robots to Cristiano Ronaldo, LeBron James, and Kobe Byrant! These are neural nets running on real hardware at our GEAR lab. Most robot demos you see online speed videos up. We actually *slow them down* so you can enjoy the fluid motions. I'm excited to announce "ASAP", a "real2sim2real" model that masters extremely smooth and dynamic motions for humanoid whole body control. We pretrain the robot in simulation first, but there is a notorious "sim2real" gap: it's very difficult for hand-engineered physics equations to match real world dynamics. Our fix is simple: just deploy a pretrained policy on real hardware, collect data, and replay the motion in sim. The replay will obviously have many errors, but that gives a rich signal to compensate for the physics discrepancy. Use another neural net to learn the delta. Basically, we "patch up" a traditional physics engine, so that the robot can experience almost the real world at scale in GPUs. The future is hybrid simulation: combine the power of classical sim engines refined over decades and the uncanny ability of modern NNs to capture a messy world.

Jim Fan

567,404 просмотров • 1 год назад

Let's reverse engineer Disney's adorable, lifelike robot! I couldn't find a whitepaper, but this is how I think it's trained: 1. The emotional behaviors are curated by Disney animation artists, keyframe by keyframe. But it cannot be "rendered" directly on the robot because it doesn't take into account the complex real-world physics. 2. Reinforcement learning (RL) is a great tool for training low-level robot controllers. RL needs a reward function to optimize, and it's typically a task reward (e.g. walk in a straight line as fast as possible). The problem is that RL doesn't know what counts as "natural behavior", and often produces weird-looking body postures that somehow still maximize the reward. This is a human alignment problem just like ChatGPT. 3. Enters Adversarial Motion Prior (AMP): a technique that learns the human preference by training a classifier on what we consider "emotional & cute". In GAN literature, this is called a discriminator. Disney artists are good at creating such a dataset. You can then add AMP as an auxiliary reward in simulation to nudge the robot towards desired behaviors. AMP was developed by Peng et al. 2021 and Escontrela et al. 2022. 4. Add lots of data augmentation to make the controller robust to physical disturbances. In RL, it's called "domain randomization". This is a very powerful technique that bridges the gap between simulator and reality. Previously, OpenAI used domain randomization to train a 5-finger robot hand to manipulate a Rubik's Cube: IEEE news article gave hints about the pipeline: Finally, praying for world peace 🙏. I hope robotics like this will bring more joy to the world.
0:58

Let's reverse engineer Disney's adorable, lifelike robot! I couldn't find a whitepaper, but this is how I think it's trained: 1. The emotional behaviors are curated by Disney animation artists, keyframe by keyframe. But it cannot be "rendered" directly on the robot because it doesn't take into account the complex real-world physics. 2. Reinforcement learning (RL) is a great tool for training low-level robot controllers. RL needs a reward function to optimize, and it's typically a task reward (e.g. walk in a straight line as fast as possible). The problem is that RL doesn't know what counts as "natural behavior", and often produces weird-looking body postures that somehow still maximize the reward. This is a human alignment problem just like ChatGPT. 3. Enters Adversarial Motion Prior (AMP): a technique that learns the human preference by training a classifier on what we consider "emotional & cute". In GAN literature, this is called a discriminator. Disney artists are good at creating such a dataset. You can then add AMP as an auxiliary reward in simulation to nudge the robot towards desired behaviors. AMP was developed by Peng et al. 2021 and Escontrela et al. 2022. 4. Add lots of data augmentation to make the controller robust to physical disturbances. In RL, it's called "domain randomization". This is a very powerful technique that bridges the gap between simulator and reality. Previously, OpenAI used domain randomization to train a 5-finger robot hand to manipulate a Rubik's Cube: IEEE news article gave hints about the pipeline: Finally, praying for world peace 🙏. I hope robotics like this will bring more joy to the world.

Jim Fan

314,694 просмотров • 3 лет назад

[RLHF] by Hand ✍️ Yesterday, Jan Leike (Jan Leike) announced he is joining #Anthropic to lead their "super-alignment" mission. He is the co-inventor of Reinforcement Learning with Human Feedback (#RLHF). How does RLHF work? [1] Given ↳ Reward Model (RM) ↳ Large Language Model (LLM) ↳ Two (Prompt, Next) Pairs 🟪 TRAIN RM Goal: Learn to give higher rewards to winners [2] Preferences ↳ A human reviews the two pairs and picks a "winner" ↳ (doc is, him) Embeddings ↳ This prompt has never received human feedback directly ↳ [S] is the special start symbol [11] Transformer ↳ Attention (yellow) ↳ Feed Forward (4x2 weight and bias matrix) ↳ Output: 3 "transformed" feature vector, one per position ↳ More details in my previous post 8. Transformer [] [12] Output Probabilities ↳ Apply a linear layer to map each transformed feature vector to a probability distribution over the vocabulary. [13] Sample ↳ Apply the greedy method, which is to pick the word with the highest score ↳ For output 1 and 2, the model accurately predicts the next word ↳ For 3rd output position, the model's predicts "him" [14] Reward Model ↳ The new pair (CEO is, him) is fed to the reward model ↳ The process is same as [3]-[6] ↳ Output: Reward = 3 [15] Loss Gradient ↳ We set the loss as the negative of the reward. ↳ The loss gradient is simply a constant -1. ↳ Run backpropagation and gradient descent to update LLM's weights and biases (red border)
1:01

[RLHF] by Hand ✍️ Yesterday, Jan Leike (Jan Leike) announced he is joining #Anthropic to lead their "super-alignment" mission. He is the co-inventor of Reinforcement Learning with Human Feedback (#RLHF). How does RLHF work? [1] Given ↳ Reward Model (RM) ↳ Large Language Model (LLM) ↳ Two (Prompt, Next) Pairs 🟪 TRAIN RM Goal: Learn to give higher rewards to winners [2] Preferences ↳ A human reviews the two pairs and picks a "winner" ↳ (doc is, him) Embeddings ↳ This prompt has never received human feedback directly ↳ [S] is the special start symbol [11] Transformer ↳ Attention (yellow) ↳ Feed Forward (4x2 weight and bias matrix) ↳ Output: 3 "transformed" feature vector, one per position ↳ More details in my previous post 8. Transformer [] [12] Output Probabilities ↳ Apply a linear layer to map each transformed feature vector to a probability distribution over the vocabulary. [13] Sample ↳ Apply the greedy method, which is to pick the word with the highest score ↳ For output 1 and 2, the model accurately predicts the next word ↳ For 3rd output position, the model's predicts "him" [14] Reward Model ↳ The new pair (CEO is, him) is fed to the reward model ↳ The process is same as [3]-[6] ↳ Output: Reward = 3 [15] Loss Gradient ↳ We set the loss as the negative of the reward. ↳ The loss gradient is simply a constant -1. ↳ Run backpropagation and gradient descent to update LLM's weights and biases (red border)

Tom Yeh

79,916 просмотров • 2 лет назад

As a Kashmiri I want to tell the whole nation that Pakistan is trying to create a big gap between kashmir and rest of India by doing terrorists attacks on innocent people in Kashmir . But we need to be united against Pakistani terrorism . If agree retweet it
0:56

As a Kashmiri I want to tell the whole nation that Pakistan is trying to create a big gap between kashmir and rest of India by doing terrorists attacks on innocent people in Kashmir . But we need to be united against Pakistani terrorism . If agree retweet it

Aquib Mir

24,811 просмотров • 1 год назад

It’s the little wins that make a big difference. I couldn’t dip this low when I started training at the beginning of the year. To see my mobility coming back and my body becoming a bit more agile is encouraging. I wanna see what I will cook by staying locked in for a whole year.
0:38

It’s the little wins that make a big difference. I couldn’t dip this low when I started training at the beginning of the year. To see my mobility coming back and my body becoming a bit more agile is encouraging. I wanna see what I will cook by staying locked in for a whole year.

Don Billiato

121,547 просмотров • 3 месяцев назад

Full Body Bench 45 Exercises in 1 bench The only workout bench you will ever need to train your whole body. CLICK ON THE LINK to see all the exercises that you can do with it. get 36kg FREE weights and a 1.2m bar only for 30,000 KES call/whatsapp 0726696926
0:53

Full Body Bench 45 Exercises in 1 bench The only workout bench you will ever need to train your whole body. CLICK ON THE LINK to see all the exercises that you can do with it. get 36kg FREE weights and a 1.2m bar only for 30,000 KES call/whatsapp 0726696926

Khaira Ummah

1,657,344 просмотров • 3 лет назад

🎙️ Excited to introduce one of my favorite projects from the past year: TeleDexter, from the BIGAI dexterity team. It’s a stable, human-level dexterous teleoperation system and a suite of autonomous policies trained with it. Pen spinning, complex in-hand reorientation, and long-horizon tool use—once seen as the holy grail of manipulation—are now unlocked. 🧵👇 The hardware is already here; we have some incredible high-DoF robotic hands. The bottleneck? The controller. Most current systems are stuck in "quasi-static" grasping mode. Meanwhile, dynamic in-hand dexterity has remained severely limited. 🧠 To unlock the massive capabilities of human-like hands, we need to build an excellent "cerebellum" for dexterous hands. TeleDexter solves this with a novel co-tracking approach: it simultaneously tracks both human hand kinematics and object states, beautifully bridging the gap between human intent and robotic control. In order to train a better co-tracking policy that works robustly in the real world, we designed : (1) a hybrid reward design that combines consecutive goal reaching and dense tracking, (2) an action masking strategy during training that enhances sim2real performance, (3) a dexterous curriculum for learning the long-horizon interactions. Each design is inspired by numerous trials and countless real-world experiments. We’ve synthesized all the system details, engineering challenges, and core insights into our latest post. If you're interested in the future of dexterous manipulation, grab a coffee and check it out (9-min read): If you have more time, check out the paper:
1:38

🎙️ Excited to introduce one of my favorite projects from the past year: TeleDexter, from the BIGAI dexterity team. It’s a stable, human-level dexterous teleoperation system and a suite of autonomous policies trained with it. Pen spinning, complex in-hand reorientation, and long-horizon tool use—once seen as the holy grail of manipulation—are now unlocked. 🧵👇 The hardware is already here; we have some incredible high-DoF robotic hands. The bottleneck? The controller. Most current systems are stuck in "quasi-static" grasping mode. Meanwhile, dynamic in-hand dexterity has remained severely limited. 🧠 To unlock the massive capabilities of human-like hands, we need to build an excellent "cerebellum" for dexterous hands. TeleDexter solves this with a novel co-tracking approach: it simultaneously tracks both human hand kinematics and object states, beautifully bridging the gap between human intent and robotic control. In order to train a better co-tracking policy that works robustly in the real world, we designed : (1) a hybrid reward design that combines consecutive goal reaching and dense tracking, (2) an action masking strategy during training that enhances sim2real performance, (3) a dexterous curriculum for learning the long-horizon interactions. Each design is inspired by numerous trials and countless real-world experiments. We’ve synthesized all the system details, engineering challenges, and core insights into our latest post. If you're interested in the future of dexterous manipulation, grab a coffee and check it out (9-min read): If you have more time, check out the paper:

Siyuan Huang

12,602 просмотров • 2 месяцев назад

New Course: Reinforcement Fine-Tuning LLMs with GRPO! Learn to use reinforcement learning to improve your LLM performance in this short course, built in collaboration with Predibase by Rubrik, and taught by Travis Addair, its Co-Founder and CTO, and Arnav Garg, its Senior Engineer and Machine Learning Lead. Reasoning models have been one of the most important developments in LLMs. Reinforcement Fine-Tuning (RFT) uses rewards to encourage LLMs to find solutions to multi-step reasoning tasks such as solving math problems and debugging code - without needing pre-existing training examples like in traditional supervised fine-tuning. Group Relative Policy Optimization (GRPO) is a reinforcement fine-tuning algorithm gaining rapid adoption. Developed by the DeepSeek team and used to train the R1 reasoning model, GRPO uses reward functions that you can write in Python to assign rewards to model responses. It’s beneficial for tasks with verifiable outcomes and can work well even with fewer than 100 training examples. It can also significantly improve the reasoning ability of smaller LLMs, making applications faster and more cost effective. In this course, you’ll take a technical deep dive into RFT with GRPO. You’ll learn to build reward functions that you can use in the GRPO training process to guide an LLM toward better performance on multi-step reasoning tasks. In detail, you’ll: - Learn when reinforcement fine-tuning is a better fit than supervised fine-tuning, especially for tasks involving multi-step reasoning or limited labeled data. - Understand how GRPO uses programmable reward functions as a more scalable alternative to the human feedback required for other reinforcement learning algorithms, such as RLHF and DPO. - Frame the Wordle game as a reinforcement fine-tuning problem and see how an LLM can learn to plan, analyze feedback, and improve its strategy over time. - Design reward functions that power the reinforcement fine-tuning process. - Learn techniques for evaluating more subjective tasks, such as rating the quality of a text summary, using an LLM as a judge. - Understand why reward hacking happens and how to avoid it by adding penalty functions to discourage undesirable behaviors. - Learn the four key components of the loss calculation in the GRPO algorithm: token probability distribution ratios, advantages, clipping, and KL-divergence. - Launch reinforcement fine-tuning jobs using Predibase’s hosted training services. By the end of this course, you’ll be able to build and fine-tune LLMs using reinforcement learning to improve reasoning without relying on large labeled datasets or subjective human feedback. Please sign up here:
2:41

New Course: Reinforcement Fine-Tuning LLMs with GRPO! Learn to use reinforcement learning to improve your LLM performance in this short course, built in collaboration with Predibase by Rubrik, and taught by Travis Addair, its Co-Founder and CTO, and Arnav Garg, its Senior Engineer and Machine Learning Lead. Reasoning models have been one of the most important developments in LLMs. Reinforcement Fine-Tuning (RFT) uses rewards to encourage LLMs to find solutions to multi-step reasoning tasks such as solving math problems and debugging code - without needing pre-existing training examples like in traditional supervised fine-tuning. Group Relative Policy Optimization (GRPO) is a reinforcement fine-tuning algorithm gaining rapid adoption. Developed by the DeepSeek team and used to train the R1 reasoning model, GRPO uses reward functions that you can write in Python to assign rewards to model responses. It’s beneficial for tasks with verifiable outcomes and can work well even with fewer than 100 training examples. It can also significantly improve the reasoning ability of smaller LLMs, making applications faster and more cost effective. In this course, you’ll take a technical deep dive into RFT with GRPO. You’ll learn to build reward functions that you can use in the GRPO training process to guide an LLM toward better performance on multi-step reasoning tasks. In detail, you’ll: - Learn when reinforcement fine-tuning is a better fit than supervised fine-tuning, especially for tasks involving multi-step reasoning or limited labeled data. - Understand how GRPO uses programmable reward functions as a more scalable alternative to the human feedback required for other reinforcement learning algorithms, such as RLHF and DPO. - Frame the Wordle game as a reinforcement fine-tuning problem and see how an LLM can learn to plan, analyze feedback, and improve its strategy over time. - Design reward functions that power the reinforcement fine-tuning process. - Learn techniques for evaluating more subjective tasks, such as rating the quality of a text summary, using an LLM as a judge. - Understand why reward hacking happens and how to avoid it by adding penalty functions to discourage undesirable behaviors. - Learn the four key components of the loss calculation in the GRPO algorithm: token probability distribution ratios, advantages, clipping, and KL-divergence. - Launch reinforcement fine-tuning jobs using Predibase’s hosted training services. By the end of this course, you’ll be able to build and fine-tune LLMs using reinforcement learning to improve reasoning without relying on large labeled datasets or subjective human feedback. Please sign up here:

Andrew Ng

86,697 просмотров • 1 год назад

Can reinforcement learning from AI feedback unlock new capabilities in AI agents? Introducing Motif, an LLM-powered method for intrinsic motivation from AI feedback. Motif extracts reward functions from Llama 2's preferences and uses them to train agents with reinforcement learning. On the complex NetHack game, Motif solves previously unsolved tasks without needing any expert demonstrations. Surprisingly, Motif's reward leads to better game score than the one obtained by using the score itself as a reward. Given access to an event captioning mechanism, a few properties make Motif a general method: • it is entirely based on open models • the LLM doesn't need direct access to the environment dynamics (e.g., its source code) • the LLM doesn't need to understand observation and action spaces The best part? You can start using Motif right now, even on a small compute budget: the whole pipeline can take less than two GPU-days. Feel free to read our paper and try our code out. Paper: Code: Blog post: Work co-lead by Martin Klissarov and myself, with Shagun Sodhani Roberta Raileanu Pierre-Luc Bacon Pascal Vincent Amy Zhang Mikael Henaff Learn more in the thread 🧵
0:52

Can reinforcement learning from AI feedback unlock new capabilities in AI agents? Introducing Motif, an LLM-powered method for intrinsic motivation from AI feedback. Motif extracts reward functions from Llama 2's preferences and uses them to train agents with reinforcement learning. On the complex NetHack game, Motif solves previously unsolved tasks without needing any expert demonstrations. Surprisingly, Motif's reward leads to better game score than the one obtained by using the score itself as a reward. Given access to an event captioning mechanism, a few properties make Motif a general method: • it is entirely based on open models • the LLM doesn't need direct access to the environment dynamics (e.g., its source code) • the LLM doesn't need to understand observation and action spaces The best part? You can start using Motif right now, even on a small compute budget: the whole pipeline can take less than two GPU-days. Feel free to read our paper and try our code out. Paper: Code: Blog post: Work co-lead by Martin Klissarov and myself, with Shagun Sodhani Roberta Raileanu Pierre-Luc Bacon Pascal Vincent Amy Zhang Mikael Henaff Learn more in the thread 🧵

Pierluca D'Oro

311,942 просмотров • 2 лет назад

Agile Continuous Jumping in Discontinuous Terrains discuss: We focus on agile, continuous, and terrain-adaptive jumping of quadrupedal robots in discontinuous terrains such as stairs and stepping stones. Unlike single-step jumping, continuous jumping requires accurately executing highly dynamic motions over long horizons, which is challenging for existing approaches. To accomplish this task, we design a hierarchical learning and control framework, which consists of a learned heightmap predictor for robust terrain perception, a reinforcement-learning-based centroidal-level motion policy for versatile and terrain-adaptive planning, and a low-level model-based leg controller for accurate motion tracking. In addition, we minimize the sim-to-real gap by accurately modeling the hardware characteristics. Our framework enables a Unitree Go1 robot to perform agile and continuous jumps on human-sized stairs and sparse stepping stones, for the first time to the best of our knowledge. In particular, the robot can cross two stair steps in each jump and completes a 3.5m long, 2.8m high, 14-step staircase in 4.5 seconds. Moreover, the same policy outperforms baselines in various other parkour tasks, such as jumping over single horizontal or vertical discontinuities.
0:25

Agile Continuous Jumping in Discontinuous Terrains discuss: We focus on agile, continuous, and terrain-adaptive jumping of quadrupedal robots in discontinuous terrains such as stairs and stepping stones. Unlike single-step jumping, continuous jumping requires accurately executing highly dynamic motions over long horizons, which is challenging for existing approaches. To accomplish this task, we design a hierarchical learning and control framework, which consists of a learned heightmap predictor for robust terrain perception, a reinforcement-learning-based centroidal-level motion policy for versatile and terrain-adaptive planning, and a low-level model-based leg controller for accurate motion tracking. In addition, we minimize the sim-to-real gap by accurately modeling the hardware characteristics. Our framework enables a Unitree Go1 robot to perform agile and continuous jumps on human-sized stairs and sparse stepping stones, for the first time to the best of our knowledge. In particular, the robot can cross two stair steps in each jump and completes a 3.5m long, 2.8m high, 14-step staircase in 4.5 seconds. Moreover, the same policy outperforms baselines in various other parkour tasks, such as jumping over single horizontal or vertical discontinuities.

AK

35,820 просмотров • 2 лет назад

This is Open Robot Parkour: the agility at Boston Dynamics level, but now open-source for everyone.🐕 Two NVIDIA tech here at play: 1. IsaacGym: train the robot dog by speeding up reality 10,000x with massively parallel simulation on GPU. 2. NVIDIA Jetson NX for onboard compute. Only vision, no LiDAR, no motion capture. Algorithm is very simple & intuitive: 1. Train in easy mode in simulation with groundtruth states. The bot quickly acquires a "sketch" of the agile skills in this step. 2. Finetune in hard mode with dynamics constraints strictly enforced. 3. Distill into a policy using realistic vision and proprioception inputs only. 4. Bridge the sim2real gap by a bit of pre-processing. The algorithm works on low-cost robot dog hardware, such as Unitree A1 and Go1. It's able to climb high obstacles, leap over large gaps, crawl under low barriers, and squeeze through thin slits. Very neat work!
0:59

This is Open Robot Parkour: the agility at Boston Dynamics level, but now open-source for everyone.🐕 Two NVIDIA tech here at play: 1. IsaacGym: train the robot dog by speeding up reality 10,000x with massively parallel simulation on GPU. 2. NVIDIA Jetson NX for onboard compute. Only vision, no LiDAR, no motion capture. Algorithm is very simple & intuitive: 1. Train in easy mode in simulation with groundtruth states. The bot quickly acquires a "sketch" of the agile skills in this step. 2. Finetune in hard mode with dynamics constraints strictly enforced. 3. Distill into a policy using realistic vision and proprioception inputs only. 4. Bridge the sim2real gap by a bit of pre-processing. The algorithm works on low-cost robot dog hardware, such as Unitree A1 and Go1. It's able to climb high obstacles, leap over large gaps, crawl under low barriers, and squeeze through thin slits. Very neat work!

Jim Fan

315,651 просмотров • 3 лет назад

How do you give a humanoid the general motion capability? Not just single motions, but all motion? Introducing SONIC, our new work on supersizing motion tracking for natural humanoid control. We argue that motion tracking is the scalable foundation task for humanoids. So we "supersized" it: 9k+ GPU hours and 100M+ motion frames. But tracking alone is not enough; we show how to make a useful control system out of it: - Universal Kinematic Planner: Enables game-like gamepad control and high-level teleoperation, just like controlling a character in a game. - VR Full-Body Teleop: Direct, real-time whole-body control by a human wearing a VR headset. - VR Keypoint Teleop: Control the upper body (hands/head) while our planner handles robust locomotion automatically. - VLA Integration: We connect this motion tracker to autonomous Visual-Language-Action (VLA) models for autonomous task execution! We use a Universal Token Space to UNIFY this command space, turning our robust tracker into a general-purpose, programmable humanoid brain. This is the generalist "System 1" for humanoids. 🚀 Project: #Humanoids #Robotics #AI #FoundationModels #NVIDIAResearch 🧠🔥
1:32

How do you give a humanoid the general motion capability? Not just single motions, but all motion? Introducing SONIC, our new work on supersizing motion tracking for natural humanoid control. We argue that motion tracking is the scalable foundation task for humanoids. So we "supersized" it: 9k+ GPU hours and 100M+ motion frames. But tracking alone is not enough; we show how to make a useful control system out of it: - Universal Kinematic Planner: Enables game-like gamepad control and high-level teleoperation, just like controlling a character in a game. - VR Full-Body Teleop: Direct, real-time whole-body control by a human wearing a VR headset. - VR Keypoint Teleop: Control the upper body (hands/head) while our planner handles robust locomotion automatically. - VLA Integration: We connect this motion tracker to autonomous Visual-Language-Action (VLA) models for autonomous task execution! We use a Universal Token Space to UNIFY this command space, turning our robust tracker into a general-purpose, programmable humanoid brain. This is the generalist "System 1" for humanoids. 🚀 Project: #Humanoids #Robotics #AI #FoundationModels #NVIDIAResearch 🧠🔥

Zhengyi “Zen” Luo

62,703 просмотров • 10 месяцев назад