Diversity-Aware Experience Replay via Statistical Leverage Scores
Diversity-Aware Experience Replay via Statistical Leverage Scores
This research introduces a diversity-aware experience replay method using statistical leverage scores to improve off-policy deep reinforcement learning in physical robots. The technique identifies rare and valuable transitions by analyzing feature column kurtosis and trajectory windows relative to goals, eliminating the need for labels or value networks. On a three-joint arm task requiring an object to be pushed around an obstacle, the approach raised success rates from 0.49 to 0.69 against an oracle ceiling of 0.74, effectively addressing the issue where replay buffers are dominated by routine data.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
success rate
0.49–0.69
Reported trials
View original evidence
n the rare situations of a log are relational, visible only in how a stretch of behavior relates to its goal, not in any single stored coordinate-per-transition scores are blind; applying the same tail rule to short trajectory windows described relative to their goal restores the signal, and on a three-joint arm that must push an object around an obstacle it raises the rare-task success from 0.49 to 0.69 against a la
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
n the rare situations of a log are relational, visible only in how a stretch of behavior relates to its goal, not in any single stored coordinate-per-transition scores are blind; applying the same tail rule to short trajectory windows described relative to their goal restores the signal, and on a three-joint arm that must push an object around an obstacle it raises the rare-task success from 0.49 to 0.69 against a la
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org