• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
  • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Chris Paxton Highlights Humanoid Manipulation World Model Research

    Chris Paxton posted about coupling global and local models for robot policies.

    CP
    1 Source, 26d ago, first seen 26d ago

    TLDR

    Chris Paxton posted a quote noting research on learning contact-rich humanoid manipulation policies purely within a world model. He stated that a very large model is usually needed to get the contact dynamics right, but the work instead couples a large global model with a local model. The post references very cool work on @RoboPapers by @Jsphamigo and @Rk4342R. It includes a 146-second video showing a silver-and-black humanoid robot.

    Combined views

    10.2K

    1 Source, first seen 26d ago

    Combined views

    10.2K

    1 Source, first seen 26d ago

    109 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    109 likes
    7 comments
    62 saves
    15 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    7 comments
    62 saves
    15 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    1 Source

    @chris_j_paxtonLearn contact-rich humanoid manipulation policies, purely within a world model. This is usually not tractable, because you need a very large model to get the contact dynamics right -- instead, @Jsphamigo @Rk4342R couple a large global model with a local model. Very cool work on @RoboPapers ->

    1 Source

    @chris_j_paxtonLearn contact-rich humanoid manipulation policies, purely within a world model. This is usually not tractable, because you need a very large model to get the contact dynamics right -- instead, @Jsphamigo @Rk4342R couple a large global model with a local model. Very cool work on @RoboPapers ->