Benchmarking and Enhancing Skill-Level Memory for Partially Observable Robotic Manipulation
HIDE benchmarks robotic memory under partial observability, and SEEK's combined memory raises task success.
HIDE is a 15-task benchmark for robotic manipulation memory under partial observability, covering repetition counting, historical-state recall, and execution-progress tracking with randomized configurations. SEEK combines three complementary memory mechanisms to retain historical evidence and track execution state. Existing policies struggle on HIDE, while the combined memory design reaches the highest average success among tested configurations in simulation and real-world experiments, though single mechanisms can hurt some tasks.
- HIDE has 15 partially observable manipulation tasks with randomized starts.
- Tasks cover counting, historical-state recall, and execution-progress tracking.
- SEEK combines three memory mechanisms to track hidden task state.
- Combined memory beats individual mechanisms on average success in sim and real robots.
From the original article
1 min read at huggingface.co
Recent advances in robot learning have enabled manipulation policies to perform increasingly diverse tasks and generalize across environments. However, reliable execution often depends on hidden task states that cannot be determined from current observations alone, making interaction history essential. We introduce HIDE, a benchmark for evaluating manipulation memory under partial observability.
By Yansong Shi, Jiange Yang, Xijie Yang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.