Not All Actions Contribute Equally: Learning Precise Progress Estimation from Imprecise Demonstrations

Key points

• Existing methods heavily rely on expert data and naively assume that all actions within a demonstration contribute to task success.1
• To this end, we propose PRIME, a novel progress estimator capable of learning precise progress estimation from imprecise demonstrations by aggregating experience across trajectories.1
• PRIME constructs a weighted graph over observed states, connecting states across trajectories based on spatial proximity and visual consistency.1
• Extensive experiments in both simulation and real world on precise manipulation tasks show that PRIME is consistently effective across diverse tasks, and robust to data with varying degrees of suboptimality.1

Sources (1)

• [1]Not All Actions Contribute Equally: Learning Precise Progress Estimation from Imprecise Demonstrations
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 05:07 PM

  Progress estimators play an essential role in providing dense reward signals for reinforcement learning in robotic manipulation, but accurately assessing task progress can be much challenging.

  Existing methods heavily rely on expert data and naively assume that all actions within a demonstration contribute to task success.

Extractive summary: sentences quoted from the sources.