Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Using reinforcement learning, we trained policies for Boston Dynamics Spot that allow the robot to achieve record running speeds of 11.5 mph (5.2 m/s) — over three times faster than Spot's default max speed.

RAI Institute

7,794 subscribers

98,538 views • 1 year ago •via X (Twitter)

Gaming Education Science & Technology

Anya Rossi• Live Now

Private livecam show

11 Comments

Loic Argelies1 year ago

@BostonDynamics Need a gallop mode…

The Rundown AI1 year ago

If you're not learning AI in 2025, you're falling behind. Join 1,000,000+ early adopters reading and learn AI in just 5 minutes a day (for free).

io.net1 year ago

@BostonDynamics At 11.5 mph, Spot is officially faster than most humans. Sleep tight.

Victor Brink1 year ago

@BostonDynamics Great! Now teach it how a change in gait like a cheetah can multiply its top speed further... 110 km/hr (70 mph) This would be epic. The Black Mirror episode featuring a running bot is "Metalhead" for further inspiration.

Elie Aljalbout1 year ago

@BostonDynamics very impressive!

TeslaElon SpaceXFan1 year ago

@BostonDynamics 🥳👍

Liberty Mint1 year ago

@BostonDynamics For sustained duration? Faster speeds achieved in sprint?

AI_TechnoKing1 year ago

@BostonDynamics Galloping gears!

Rodomonte1 year ago

@BostonDynamics @AskPerplexity give me calculations on max possible physical speed for a robot of that weight

Pat Dunne1 year ago

@BostonDynamics Yup RL.. just like motors.. it’s a thing .. welcome BD back to the race

Tom Ramirez1 year ago

@BostonDynamics Robot PT

Related Videos

Reinforcement learning is used to speed the production of behavior for the Boston Dynamics Atlas humanoid robot. At the heart of the learning process is a physics-based simulator that generates training data for a variety of maneuvers.

Reinforcement learning is used to speed the production of behavior for the Boston Dynamics Atlas humanoid robot. At the heart of the learning process is a physics-based simulator that generates training data for a variety of maneuvers.

RAI Institute

76,595 views • 1 year ago

1. Unitree B2 W– Military & Police A quadruped robot faster than Boston Dynamics’ Spot. Used for surveillance, tactical drills, and even fire rescue. Top speed: 30 m/s the world’s fastest robot dog. maximum weight can carry 40KG

1. Unitree B2 W– Military & Police A quadruped robot faster than Boston Dynamics’ Spot. Used for surveillance, tactical drills, and even fire rescue. Top speed: 30 m/s the world’s fastest robot dog. maximum weight can carry 40KG

ZOYA ✪

60,475 views • 1 year ago

Watch Spot crouch, jump, climb boxes and leap across gaps, controlled by a neural network trained with reinforcement learning (RL) and multi-expert distillation. Multiple expert policies were trained and distilled together into a single policy that was fine tuned to improve performance over diverse terrains. This work was inspired by ANYmal’s parkour capabilities. The neural network processes depth data from Spot's sensors to construct an understanding of the environment.

Watch Spot crouch, jump, climb boxes and leap across gaps, controlled by a neural network trained with reinforcement learning (RL) and multi-expert distillation. Multiple expert policies were trained and distilled together into a single policy that was fine tuned to improve performance over diverse terrains. This work was inspired by ANYmal’s parkour capabilities. The neural network processes depth data from Spot's sensors to construct an understanding of the environment.

RAI Institute

14,768 views • 2 months ago

RoboBall is a spherical robot developed by researchers at Texas A&M University to explore lunar craters. It moves using a pendulum system inside the ball, allowing it to roll and steer in any direction. The robot can reach speeds of up to 20 mph, and researchers are working to make it faster and more autonomous for future space missions.

RoboBall is a spherical robot developed by researchers at Texas A&M University to explore lunar craters. It moves using a pendulum system inside the ball, allowing it to roll and steer in any direction. The robot can reach speeds of up to 20 mph, and researchers are working to make it faster and more autonomous for future space missions.

Space and Technology

20,739 views • 1 month ago

Robot Learning needs 4D world models! Robot Learning needs 4D world models! Robot Learning needs 4D world models! We introduce TesserAct, a 4D embodied world model that can simulate how agents interact with the 3D world over time! We achieve this by simply extending a pre-trained 2D video generation model to jointly predict RGB, depth, and surface normals. It enables: 1️⃣ Much better policy learning in the wild 2️⃣ Temporal + spatial coherence in 4D dynamic prediction 3️⃣ Novel view synthesis for embodied scenes Code: Paper Link: Project page:

Robot Learning needs 4D world models! Robot Learning needs 4D world models! Robot Learning needs 4D world models! We introduce TesserAct, a 4D embodied world model that can simulate how agents interact with the 3D world over time! We achieve this by simply extending a pre-trained 2D video generation model to jointly predict RGB, depth, and surface normals. It enables: 1️⃣ Much better policy learning in the wild 2️⃣ Temporal + spatial coherence in 4D dynamic prediction 3️⃣ Novel view synthesis for embodied scenes Code: Paper Link: Project page:

Chuang Gan

43,265 views • 1 year ago

Over the last few months, we’ve been thinking about how to learn from “off-domain” data - data from non-robot sources like video or simulation. These data sources are not quite good enough to learn policies (even monolithic VLA models) directly, but they still contain lots of information that can be useful for generalizable robot control. How can we develop robot learning models that are able to make use of this type of data for generalizable control? In new work, that we call HAMSTER, we show that VLMs can be useful for enabling robotic learning from off-domain data, but specifically when used through hierarchical VLA architectures. We show that this class of models can learn generalizable robot policies for the real world from large-scale, off-domain data. A 🧵 (1/10)

Over the last few months, we’ve been thinking about how to learn from “off-domain” data - data from non-robot sources like video or simulation. These data sources are not quite good enough to learn policies (even monolithic VLA models) directly, but they still contain lots of information that can be useful for generalizable robot control. How can we develop robot learning models that are able to make use of this type of data for generalizable control? In new work, that we call HAMSTER, we show that VLMs can be useful for enabling robotic learning from off-domain data, but specifically when used through hierarchical VLA architectures. We show that this class of models can learn generalizable robot policies for the real world from large-scale, off-domain data. A 🧵 (1/10)

Abhishek Gupta

11,994 views • 1 year ago

Policies trained on real robot data via imitation can be surprisingly capable. But for domains like dexterous manipulation, they are often not quite good enough: they move slowly, miss grasps, make unreliable contact, and fail under small perturbations. Can we improve them without any additional data collection on the real robot? In SCORE, we show that we can improve real-world diffusion/flow policies cheaply by using simulation to simply learn how to steer them on deployment. This leads to large gains in real-world success and speed across a variety of tasks, without requiring additional real-world experience: 🧵 (1/10)

Policies trained on real robot data via imitation can be surprisingly capable. But for domains like dexterous manipulation, they are often not quite good enough: they move slowly, miss grasps, make unreliable contact, and fail under small perturbations. Can we improve them without any additional data collection on the real robot? In SCORE, we show that we can improve real-world diffusion/flow policies cheaply by using simulation to simply learn how to steer them on deployment. This leads to large gains in real-world success and speed across a variety of tasks, without requiring additional real-world experience: 🧵 (1/10)

Abhishek Gupta

33,455 views • 17 days ago

$CHAMELEONS SNAPS FOOD FASTER THAN A SPEEDING CAR This girl hand-feeds her veiled chameleon, a species famous for color-changing skin and super-speedy tongues that can rocket out up to twice their body length to snag prey. Chameleons’ tongues accelerate faster than a fighter jet, hitting speeds over 13 mph in just fractions of a second. Sprinkles the chameleon nails the strike every time - proof that cuteness and ninja-level hunting skills definitely coexist. Source: sprinklesthecham$

CHAMELEONS SNAPS FOOD FASTER THAN A SPEEDING CAR This girl hand-feeds her veiled chameleon, a species famous for color-changing skin and super-speedy tongues that can rocket out up to twice their body length to snag prey. Chameleons’ tongues accelerate faster than a fighter jet, hitting speeds over 13 mph in just fractions of a second. Sprinkles the chameleon nails the strike every time - proof that cuteness and ninja-level hunting skills definitely coexist. Source: sprinklesthecham

Mario Nawfal

50,556 views • 1 year ago

Teleoperating a robot is hard. This means that when performing a robot task via teleoperation — say, to collect examples for training a robot policy — it’s almost unavoidably slower than you would like, below either the capabilities of the human expert on their own or the robot performing the task. Wouldn’t it be great if there was a way to fix this? Unfortunately, it’s harder than it looks. You can’t just execute faster, as this alters the distribution of environment states the policy will encounter. Nadun Ranawaka Arachchige and Zhenyang Chen propose Speed-Adaptive Imitation Learning (SAIL), which adds error-adaptive guidance, adapts execution speed according to task structure, predicts controller-invariant action targets to ensure robustness across execution speeds, and explicitly models delays from, for example, sensor latency. Watch episode #59 of RoboPapers, with Michael Cho - Rbt/Acc and Chris Paxton to learn more!

Teleoperating a robot is hard. This means that when performing a robot task via teleoperation — say, to collect examples for training a robot policy — it’s almost unavoidably slower than you would like, below either the capabilities of the human expert on their own or the robot performing the task. Wouldn’t it be great if there was a way to fix this? Unfortunately, it’s harder than it looks. You can’t just execute faster, as this alters the distribution of environment states the policy will encounter. Nadun Ranawaka Arachchige and Zhenyang Chen propose Speed-Adaptive Imitation Learning (SAIL), which adds error-adaptive guidance, adapts execution speed according to task structure, predicts controller-invariant action targets to ensure robustness across execution speeds, and explicitly models delays from, for example, sensor latency. Watch episode #59 of RoboPapers, with Michael Cho - Rbt/Acc and Chris Paxton to learn more!

RoboPapers

27,970 views • 5 months ago

Senator JD Vance: "For three years Kamala Harris went out bragging that she was going to undo Donald Trump's border policies. She did exactly that. We had a record number of illegal crossings. We had a record number of fentanyl coming into our country, and now that she's running for President she says she's got religion...The only thing that she did when she became the Border Czar was to undo 94 Donald Trump executive actions to open up southern border."

Senator JD Vance: "For three years Kamala Harris went out bragging that she was going to undo Donald Trump's border policies. She did exactly that. We had a record number of illegal crossings. We had a record number of fentanyl coming into our country, and now that she's running for President she says she's got religion...The only thing that she did when she became the Border Czar was to undo 94 Donald Trump executive actions to open up southern border."

JD Vance

2,207,340 views • 1 year ago

Israel-based Mentee Robotics has demonstrated a logistics workflow: two MenteeBot V3 humanoids work autonomously to pick and place totes. A Modular Agent System is preferred because it favors real-world robustness and lower compute needs over the End-to-End VLA model. Its architecture is composed of three components: - LLM Planner: Converts instructions into executable Robotic API Language code for reliable task decomposition and error handling. - Perception Stack: Uses pre-trained models (NeRF/3DGS, distilled vision) for scene understanding and navigation. - Control Policies: Reinforcement Learning (RL) models, trained at scale via Sim2Real, generate motor commands, enabling high-accuracy mobile manipulation. Crucially, the robot learns new tasks from a single demonstration in hours. Object tracking uses 3D geometry (STL/URDF) tracked in the video to define the RL reward function. Training is optimized using 'Automatic Curriculum Learning', which autonomously adjusts task difficulty based on robot performance, eliminating manual engineering. All computation runs onboard.

Israel-based Mentee Robotics has demonstrated a logistics workflow: two MenteeBot V3 humanoids work autonomously to pick and place totes. A Modular Agent System is preferred because it favors real-world robustness and lower compute needs over the End-to-End VLA model. Its architecture is composed of three components: - LLM Planner: Converts instructions into executable Robotic API Language code for reliable task decomposition and error handling. - Perception Stack: Uses pre-trained models (NeRF/3DGS, distilled vision) for scene understanding and navigation. - Control Policies: Reinforcement Learning (RL) models, trained at scale via Sim2Real, generate motor commands, enabling high-accuracy mobile manipulation. Crucially, the robot learns new tasks from a single demonstration in hours. Object tracking uses 3D geometry (STL/URDF) tracked in the video to define the RL reward function. Training is optimized using 'Automatic Curriculum Learning', which autonomously adjusts task difficulty based on robot performance, eliminating manual engineering. All computation runs onboard.

The Humanoid Hub

15,729 views • 7 months ago

$Boston Dynamics’ Spot costs $75,000. This one? $135. LeCabot is an open-source mod for your SO-100 and Unitree Go2 that turns them into a fully capable mobile manipulator… for a fraction of the price. You control both the robot dog and arm simultaneously using a Meta Quest headset, thanks to Phospho AI 🧪 It’s real, it’s accessible, and it’s ready for AI model training on rough terrain. Start building: Bill of materials and assembly instructions →$

Boston Dynamics’ Spot costs $75,000. This one? $135. LeCabot is an open-source mod for your SO-100 and Unitree Go2 that turns them into a fully capable mobile manipulator… for a fraction of the price. You control both the robot dog and arm simultaneously using a Meta Quest headset, thanks to Phospho AI 🧪 It’s real, it’s accessible, and it’s ready for AI model training on rough terrain. Start building: Bill of materials and assembly instructions →

Ilir Aliu - eu/acc

20,614 views • 10 months ago

“The KPI metric I’d focus on would be 87%+ max sprint effort and distance. This would show our repeat sprint ability at high speeds and if we are able to perform those efforts for a longer amount of time. By using these in combination it would allow us to identify their ability to translate winter training speed potential and optimise their training stimulus going forward.“ TCU Football’s Tyrell Brown (Pop Pop) ✌ By using our fully customisable speed zones, practitioners can set speed zones that are relative to a player’s max speed, making it extremely easy for coaches to identify the 87%+ speed exposures across their whole squad. Additionally, the distance covered in these relative speed zones can also be analysed at the click of a button allowing both the effort count and volume of sprints to be accounted for in player monitoring. This can also be achieved by setting the players sprint entry speed to 87%, dependent on user preference. Link in the bio to read more on this.

“The KPI metric I’d focus on would be 87%+ max sprint effort and distance. This would show our repeat sprint ability at high speeds and if we are able to perform those efforts for a longer amount of time. By using these in combination it would allow us to identify their ability to translate winter training speed potential and optimise their training stimulus going forward.“ TCU Football’s Tyrell Brown (Pop Pop) ✌ By using our fully customisable speed zones, practitioners can set speed zones that are relative to a player’s max speed, making it extremely easy for coaches to identify the 87%+ speed exposures across their whole squad. Additionally, the distance covered in these relative speed zones can also be analysed at the click of a button allowing both the effort count and volume of sprints to be accounted for in player monitoring. This can also be achieved by setting the players sprint entry speed to 87%, dependent on user preference. Link in the bio to read more on this.

STATSports

10,244 views • 1 year ago

Today, we're joined by Nikita Rudin, co-founder and CEO of Flexion to discuss the gap between current robotic capabilities and what’s required to deploy fully autonomous robots in the real world. Nikita explains how reinforcement learning and simulation have driven rapid progress in robot locomotion—and why locomotion is still far from “solved.” We dig into the sim2real gap, and how adding visual inputs introduces noise and significantly complicates sim-to-real transfer. We also explore the debate between end-to-end models and modular approaches, and why separating locomotion, planning, and semantics remains a pragmatic approach today. Nikita also introduces the concept of "real-to-sim", which uses real-world data to refine simulation parameters for higher fidelity training, discusses how reinforcement learning, imitation learning, and teleoperation data are combined to train robust policies for both quadruped and humanoid robots, and introduces Flexion's hierarchical approach that utilizes pre-trained Vision-Language Models (VLMs) for high-level task orchestration with Vision-Language-Action (VLA) models and low-level whole-body trackers. Finally, Nikita shares the behind-the-scenes in humanoid robot demos, his take on reinforcement learning in simulation versus the real world, the nuances of reward tuning, and offers practical advice for researchers and practitioners looking to get started in robotics today. 🗒️ For the full list of resources for this episode, visit the show notes page: 📖 CHAPTERS =============================== 00:00 - Introduction 04:07 - Is robot locomotion solved? 06:04 - Sim-to-real gap 08:58 - Adding semantics to policies 09:42 - Modular vs end-to-end architectures 10:29 - Planner model 12:21 - Adapting RL techniques from quadrupeds to humanoids 15:39 - Behind robot demos 18:09 - Humanoid robots in home environments 22:03 - Training approach 23:56 - VLA models 27:59 - Closing the sim-to-real gap 32:55 - Task orchestration using VLMs 36:38 - Tool use 38:10 - Model hierarchy 43:37 - Simulator versus simulation environment 44:57 - Combining imitation learning and reinforcement learning 46:42 - RL in real world versus RL in simulation 52:58 - Reward tuning and value functions in robotics 56:38 - Predictions 1:00:10 - Humanoids, quadropeds, and wheeled platforms 1:02:45 - Advice, recommended robot kits, and community pla

Today, we're joined by Nikita Rudin, co-founder and CEO of Flexion to discuss the gap between current robotic capabilities and what’s required to deploy fully autonomous robots in the real world. Nikita explains how reinforcement learning and simulation have driven rapid progress in robot locomotion—and why locomotion is still far from “solved.” We dig into the sim2real gap, and how adding visual inputs introduces noise and significantly complicates sim-to-real transfer. We also explore the debate between end-to-end models and modular approaches, and why separating locomotion, planning, and semantics remains a pragmatic approach today. Nikita also introduces the concept of "real-to-sim", which uses real-world data to refine simulation parameters for higher fidelity training, discusses how reinforcement learning, imitation learning, and teleoperation data are combined to train robust policies for both quadruped and humanoid robots, and introduces Flexion's hierarchical approach that utilizes pre-trained Vision-Language Models (VLMs) for high-level task orchestration with Vision-Language-Action (VLA) models and low-level whole-body trackers. Finally, Nikita shares the behind-the-scenes in humanoid robot demos, his take on reinforcement learning in simulation versus the real world, the nuances of reward tuning, and offers practical advice for researchers and practitioners looking to get started in robotics today. 🗒️ For the full list of resources for this episode, visit the show notes page: 📖 CHAPTERS =============================== 00:00 - Introduction 04:07 - Is robot locomotion solved? 06:04 - Sim-to-real gap 08:58 - Adding semantics to policies 09:42 - Modular vs end-to-end architectures 10:29 - Planner model 12:21 - Adapting RL techniques from quadrupeds to humanoids 15:39 - Behind robot demos 18:09 - Humanoid robots in home environments 22:03 - Training approach 23:56 - VLA models 27:59 - Closing the sim-to-real gap 32:55 - Task orchestration using VLMs 36:38 - Tool use 38:10 - Model hierarchy 43:37 - Simulator versus simulation environment 44:57 - Combining imitation learning and reinforcement learning 46:42 - RL in real world versus RL in simulation 52:58 - Reward tuning and value functions in robotics 56:38 - Predictions 1:00:10 - Humanoids, quadropeds, and wheeled platforms 1:02:45 - Advice, recommended robot kits, and community pla

The TWIML AI Podcast

22,533 views • 6 months ago

$Boston Dynamics’ Spot costs $75,000. This one? $135. LeCabot is an open-source mod for your SO-100 and Unitree Go2 that turns them into a fully capable mobile manipulator… for a fraction of the price. You control both the robot dog and arm simultaneously using a Meta Quest headset, thanks to phospho 🧪 🧪 It’s real, it’s accessible, and it’s ready for AI model training on rough terrain. Thanks for sharing, ulyanov! Start building: Bill of materials and assembly instructions →$

Boston Dynamics’ Spot costs $75,000. This one? $135. LeCabot is an open-source mod for your SO-100 and Unitree Go2 that turns them into a fully capable mobile manipulator… for a fraction of the price. You control both the robot dog and arm simultaneously using a Meta Quest headset, thanks to phospho 🧪 🧪 It’s real, it’s accessible, and it’s ready for AI model training on rough terrain. Thanks for sharing, ulyanov! Start building: Bill of materials and assembly instructions →

Ilir Aliu - eu/acc

21,874 views • 1 year ago

President Trump hypes up the massive, high-speed racing event set to take over the nation's capital on August 22nd and 23rd, promising a historic spectacle down Pennsylvania Avenue to celebrate America's 250th anniversary. The event will feature cars reaching speeds of over 190 mph racing around the National Mall. "It's going to be a sight for the ages. It really will be something special. And they're going to go all around the National Mall at speeds that are not quite- if we did that during the regular day, I think people would be put in jail, perhaps for the rest of their lives."

President Trump hypes up the massive, high-speed racing event set to take over the nation's capital on August 22nd and 23rd, promising a historic spectacle down Pennsylvania Avenue to celebrate America's 250th anniversary. The event will feature cars reaching speeds of over 190 mph racing around the National Mall. "It's going to be a sight for the ages. It really will be something special. And they're going to go all around the National Mall at speeds that are not quite- if we did that during the regular day, I think people would be put in jail, perhaps for the rest of their lives."

Fox News

137,347 views • 6 days ago

NASA took a major step toward redefining high-speed flight on Nov. 16, 2004. The X-43A, a small, experimental aircraft, proved that an engine drawing oxygen from the atmosphere could propel an aircraft to hypersonic speeds – greater than Mach 5, or five times the speed of sound. During the flight over the Pacific Ocean west of California, the X-43A set the world speed record for a jet-powered aircraft, reaching Mach 9.6, or nearly 7,000 mph. The mission originated from NASA’s Armstrong Flight Research Center in Edwards, California. The X-43A flight showed future high-speed vehicles could have more airplane-like operations for increased affordability, flexibility, and safety. NASA sees such technology as a key to unlocking improved access to space and future routine flight across the world. Explore more about hypersonic research at NASA Armstrong:

NASA took a major step toward redefining high-speed flight on Nov. 16, 2004. The X-43A, a small, experimental aircraft, proved that an engine drawing oxygen from the atmosphere could propel an aircraft to hypersonic speeds – greater than Mach 5, or five times the speed of sound. During the flight over the Pacific Ocean west of California, the X-43A set the world speed record for a jet-powered aircraft, reaching Mach 9.6, or nearly 7,000 mph. The mission originated from NASA’s Armstrong Flight Research Center in Edwards, California. The X-43A flight showed future high-speed vehicles could have more airplane-like operations for increased affordability, flexibility, and safety. NASA sees such technology as a key to unlocking improved access to space and future routine flight across the world. Explore more about hypersonic research at NASA Armstrong:

NASA Armstrong

61,435 views • 8 months ago

Tired of teleoperation? One human video → 1,000s of robot demos. (📍GitHub ) Scaling Robot Data Without Dynamics Simulation or Robot Hardware Real2Render2Real (R2R2R) is a new way to scale robot data without physics simulation or hardware. You take a phone scan + a single monocular human demo. It tracks the motion, renders photorealistic scenes, and generates diverse, robot-agnostic trajectories ready for training. > No teleop, no sim, no robot, just a phone and a video > Train VLA models and diffusion policies directly on the output > Supports multiple robot embodiments with kinematic consistency > 1000s of demos in 1/27 the time of real-world collection Thank you, Max Fu, for sharing!! Project: Paper: Code coming soon: It shows that with the right pipeline, you can scale robot learning data without touching a robot. One of the most interesting directions in scalable robotics today. —— Weekly robotics and AI insights. Subscribe free:

Tired of teleoperation? One human video → 1,000s of robot demos. (📍GitHub ) Scaling Robot Data Without Dynamics Simulation or Robot Hardware Real2Render2Real (R2R2R) is a new way to scale robot data without physics simulation or hardware. You take a phone scan + a single monocular human demo. It tracks the motion, renders photorealistic scenes, and generates diverse, robot-agnostic trajectories ready for training. > No teleop, no sim, no robot, just a phone and a video > Train VLA models and diffusion policies directly on the output > Supports multiple robot embodiments with kinematic consistency > 1000s of demos in 1/27 the time of real-world collection Thank you, Max Fu, for sharing!! Project: Paper: Code coming soon: It shows that with the right pipeline, you can scale robot learning data without touching a robot. One of the most interesting directions in scalable robotics today. —— Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

42,804 views • 5 months ago