The authors attached lightweight actor and critic networks to pretrained robot foundation models, including vision-language-action and world action models.
ACCRUE helps robots learn new assembly tasks without forgetting old ones
The framework carries compact actor and critic networks between tasks, improving learning speed while preserving earlier behaviors through limited replay and distillation.
Big Tech
Changyeon Kim · Kimin Lee · Yashraj Narang · Yijie Guo
NVIDIA · KAIST
Research Digest··3 min read
Kim and colleagues present ACCRUE, an off-policy reinforcement learning framework for continually adapting robot foundation models to new tasks.
Why this paper
From NVIDIA and KAIST
In one line
ACCRUE transfers lightweight actor-critic networks across tasks for efficient continual off-policy RL with minimal forgetting.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§