Skip navigation
Diffusion and flow matching-based robot planners are slow and generate noisy and jerky trajectories.
0:00 / 0:23

Diffusion and flow matching-based robot planners are slow and generate noisy and jerky trajectories.

Ke Li 🍁

@KL_Div7.4K followers

Visit on X
796 likesDownloadOriginal post

102,968 views

Diffusion and flow matching-based robot planners are slow and generate noisy and jerky trajectories. Delighted to share our ICRA 2026 paper, which leverages IMLE to improve planning frequency 19-fold from 4.3 Hz to 83 Hz and reduces jerk by 38% relative to flow matching. Joint work w/ Grayson Lee, Minh Bui, Shuzi Zhou, Yankai Li and Mo Chen. (1/7)

16 Comments

Comments from the original post on X (Twitter)

  • Ke Li 🍁 @ ECCV 2026's profile picture

    @KL_Div

    Generative models, such as diffusion models and flow matching, are widely used for robot planning. The goal is to model the probability distribution over plausible future trajectories conditioned on the initial state. (2/7)

    21 likes

  • Ke Li 🍁 @ ECCV 2026's profile picture

    @KL_Div

    Our method uses Implicit Maximum Likelihood Estimation (IMLE) to train a one-step generative model on trajectories of varying quality. IMLE works by having each trajectory in the training dataset pull the closest generated trajectory towards it. (3/7)

    13 likes

  • Ke Li 🍁 @ ECCV 2026's profile picture

    @KL_Div

    Vanilla IMLE treats all trajectories in the training dataset equally despite their varying quality. Since our goal is to generate trajectories of high quality, we propose reward-weighted IMLE, which emphasizes trajectories with high reward more than those with low reward. (4/7)

    8 likes

  • Ke Li 🍁 @ ECCV 2026's profile picture

    @KL_Div

    We demonstrate on offline reinforcement learning tasks that our IMLE model runs in real time on both the CPU and GPU and achieves an order-of-magnitude speedup relative to diffusion models. (5/7)

    8 likes

  • Ke Li 🍁 @ ECCV 2026's profile picture

    @KL_Div

    We deploy our planner onboard a mobile robot and demonstrate real-time navigation in dynamic multi-agent environments. (6/7)

    8 likes

  • Ke Li 🍁 @ ECCV 2026's profile picture

    @KL_Div

    For more details, see: - Project website: - Paper: - Code: If you are at ICRA, come check out Poster 301 on Tuesday afternoon! (7/7)

    24 likes

  • Mr V's profile picture

    @InvntorV

    Bruh diffusion models are really bad at generalizing in real-time , learnt a lot from your paper and I think I will find myself always going back to this. Who would have thought that the denoising process is the root cause. My initial thoughts were that without the de-noisers generalizations wouldn’t work or be optimal. This is great works and it is real time.

    2 likes

  • Emeka's profile picture

    @EmekaBuilds_

    This is so nice ! 👏👏👏

    1 like

  • oriel haim's profile picture

    @orielhaim

    This is really cool! Are you looking for employees? Not for the salary I just like to play with things like this 🙃

  • Julley Thai's profile picture

    @julleybuilds

    Lil bro playing crossing road.

  • Priyav K Kaneria's profile picture

    @_diginova

    really cool

  • Mooncast Productions's profile picture

    @MooncastOnline

    boids demo?

  • Alper Ahmetoglu's profile picture

    @alperahmetoglu

    Felt nostalgic. 2018, I was trying something like IMLE in the other KL direction, and was trying to find a way to prevent collapse. Then I stumbled upon IMLE. Cool idea! Will check this out.

  • LSMO's profile picture

    @LSbyMO

    Cool

  • Saleh Aldwais's profile picture

    @SAldwais

    This is good it didn’t stop at all butterfly find a way while moving

  • Wendy Carlosa's profile picture

    @WendyCarlosa

    looks like how the military would operate if the soldiers were cats