Sample-Efficient Reinforcement Learning for Policy Improvement via Flow-Reversed Action Anchors
Sample-Efficient Reinforcement Learning for Policy Improvement via Flow-Reversed Action Anchors
This research presents AnchorQ, a method that enhances generalist robot policies through structured latent action representations and flow reversal. By mapping reference behaviors to state-conditioned latent anchors and applying value-guided sampling, AnchorQ achieves significant improvements in sample efficiency, showing a 37% increase in real-world success rate and 97% higher simulated task throughput compared to existing baselines.
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
Reported success rate
97%
View original evidence
asks, AnchorQ outperforms state-of-the-art post-training baselines in sample efficiency, delivering a 37% improvement in real-world success rate and 97% higher simulated task throughput under the same interaction budget. Additional visualizations are available on our website https://anchorq-rl.com/.
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.
asks, AnchorQ outperforms state-of-the-art post-training baselines in sample efficiency, delivering a 37% improvement in real-world success rate and 97% higher simulated task throughput under the same interaction budget. Additional visualizations are available on our website https://anchorq-rl.com/.
Open source S5
Sample-Efficient Reinforcement Learning for Policy Improvement via Flow-Reversed Action Anchors
Open source S1
Source：arXiv Robotics — research abstracts · arxiv.org