PREM: Prediction-Regulated Memory for Robot Manipulation
PREM: Prediction-Regulated Memory for Robot Manipulation
This research introduces Prediction-Regulated Memory (PREM), a framework that leverages action-conditioned future prediction to guide memory learning and updating in vision-language-action policies.
The paper introduces Prediction-Regulated Memory (PREM), a framework that uses action-conditioned future prediction to regulate visual memory updates in vision-language-action policies, demonstrating improved success rates on DOMINO and RMBench benchmarks with a reported inference overhead of 7.7 ms on a single NVIDIA H800 GPU.
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
20.74%
View original evidence
each spatial location. The updated visual state supplies temporal context for action generation alongside current observations. Built on $\pi_{0.5}$, PREM improves average success rate from 9.63% to 20.74% on DOMINO dynamic manipulation tasks and from 14.4% to 83.8% on the RMBench M(1) memory-dependent setting. Further evaluations on LIBERO and LIBERO-Pro show improvements in general manipulation and robustness to s
Open source S4Reported success rate
83.8%
View original evidence
each spatial location. The updated visual state supplies temporal context for action generation alongside current observations. Built on $\pi_{0.5}$, PREM improves average success rate from 9.63% to 20.74% on DOMINO dynamic manipulation tasks and from 14.4% to 83.8% on the RMBench M(1) memory-dependent setting. Further evaluations on LIBERO and LIBERO-Pro show improvements in general manipulation and robustness to s
Open source S4mean inference overhead
7.7
View original evidence
patial perturbations. PREM incurs a mean inference overhead of 7.7 ms on a single NVIDIA H800 GPU.
Open source S5Reported success rate
9.63%
View original evidence
each spatial location. The updated visual state supplies temporal context for action generation alongside current observations. Built on $\pi_{0.5}$, PREM improves average success rate from 9.63% to 20.74% on DOMINO dynamic manipulation tasks and from 14.4% to 83.8% on the RMBench M(1) memory-dependent setting. Further evaluations on LIBERO and LIBERO-Pro show improvements in general manipulation and robustness to s
Open source S4Reported success rate
14.4%
View original evidence
each spatial location. The updated visual state supplies temporal context for action generation alongside current observations. Built on $\pi_{0.5}$, PREM improves average success rate from 9.63% to 20.74% on DOMINO dynamic manipulation tasks and from 14.4% to 83.8% on the RMBench M(1) memory-dependent setting. Further evaluations on LIBERO and LIBERO-Pro show improvements in general manipulation and robustness to s
Open source S4
- paper: 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.
each spatial location. The updated visual state supplies temporal context for action generation alongside current observations. Built on $\pi_{0.5}$, PREM improves average success rate from 9.63% to 20.74% on DOMINO dynamic manipulation tasks and from 14.4% to 83.8% on the RMBench M(1) memory-dependent setting. Further evaluations on LIBERO and LIBERO-Pro show improvements in general manipulation and robustness to s
Open source S4
patial perturbations. PREM incurs a mean inference overhead of 7.7 ms on a single NVIDIA H800 GPU.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org