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

FIND reportedly raises robot task success from 55% to 71.9% across eight tasks

A post sharing the preprint says FIND selects feasible tasks, targets weaker skills and evaluates its own attempts.

1 Source, 49m ago, first seen 49m ago

TLDR

FIND lets a robot guide its practice and learn from real-world experience, according to a post sharing the new preprint. The poster reports success across eight tasks rising from 55% to 71.9%. They also say FIND completed 456 episodes in a representative six-hour run, with 30 human scene-recovery interventions and no human-provided reward labels.

Combined views

—

1 Source, first seen 49m ago

— likes— comments— saves— reposts

Combined views

—

1 Source, first seen 49m ago

— likes— comments— saves— reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

1 Source

Georgia Chalvatzaki@GeorgiaChalA robot that identifies its weaknesses, decides what to practice, and gets better through real-world experience. 🤖🔁 Steven and Yuan built FIND to close this loop: the robot selects feasible tasks, focuses on weaker skills, evaluates its own attempts, and improves its policy through real-world reinforcement learning. And the practice pays off: success across eight tasks rises from 55% to 71.9%. In a representative six-hour run, FIND completes 456 episodes with just 30 human scene-recovery interventions and no human-provided reward labels. Very excited to share our new preprint! A step toward robots that take charge of their own learning.49m
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    1 Source

    Georgia Chalvatzaki@GeorgiaChalA robot that identifies its weaknesses, decides what to practice, and gets better through real-world experience. 🤖🔁 Steven and Yuan built FIND to close this loop: the robot selects feasible tasks, focuses on weaker skills, evaluates its own attempts, and improves its policy through real-world reinforcement learning. And the practice pays off: success across eight tasks rises from 55% to 71.9%. In a representative six-hour run, FIND completes 456 episodes with just 30 human scene-recovery interventions and no human-provided reward labels. Very excited to share our new preprint! A step toward robots that take charge of their own learning.49m
    Today's Rank

    #8

    Today's Rank

    #8