Correct Robots Before They Make Mistakes: Proactive Human-in-the-Loop Intervention via Preference Learning
Correct Robots Before They Make Mistakes: Proactive Human-in-the-Loop Intervention via Preference Learning
This paper introduces PHIL, a framework enabling operators to preview and intervene on robot policy predictions before undesirable actions occur. Using an XR interface, PHIL displays predicted trajectories during takeover and pairs rejected predictions with corrective action chunks for preference-based post-training. Evaluated on four real-world manipulation tasks with five correction episodes per condition, PHIL increases mean success rates from 44.0% to 70.3% for diffusion policy and from 57.8% to 76.9% for X-VLA compared to HG-DAgger, while reducing unnecessary recovery behaviors in successful episodes.
The study evaluates a proactive human-in-the-loop framework on four real-world manipulation tasks, reporting success rate improvements for diffusion policy and X-VLA models compared to HG-DAgger baseline.
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.
- Control
- Shared controlOpen source S2
Reported numbers
Reported success rate
44%
Reported trials
View original evidence
e correction episodes per condition, PHIL increases mean success rates across the evaluated tasks and demonstration budgets from 44.0% to 70.3% for the diffusion policy and from 57.8% to 76.9% for X-VLA, compared with HG-DAgger. Separately, policies trained with proactive corrections show fewer unnecessary recovery behaviors in successful episodes. Project page: https://intuitive-robots.github.io/PHIL-page/.
Open source S4Reported success rate
70.3%
Reported trials
View original evidence
e correction episodes per condition, PHIL increases mean success rates across the evaluated tasks and demonstration budgets from 44.0% to 70.3% for the diffusion policy and from 57.8% to 76.9% for X-VLA, compared with HG-DAgger. Separately, policies trained with proactive corrections show fewer unnecessary recovery behaviors in successful episodes. Project page: https://intuitive-robots.github.io/PHIL-page/.
Open source S4Reported success rate
57.8%
Reported trials
View original evidence
e correction episodes per condition, PHIL increases mean success rates across the evaluated tasks and demonstration budgets from 44.0% to 70.3% for the diffusion policy and from 57.8% to 76.9% for X-VLA, compared with HG-DAgger. Separately, policies trained with proactive corrections show fewer unnecessary recovery behaviors in successful episodes. Project page: https://intuitive-robots.github.io/PHIL-page/.
Open source S4Reported success rate
76.9%
Reported trials
View original evidence
e correction episodes per condition, PHIL increases mean success rates across the evaluated tasks and demonstration budgets from 44.0% to 70.3% for the diffusion policy and from 57.8% to 76.9% for X-VLA, compared with HG-DAgger. Separately, policies trained with proactive corrections show fewer unnecessary recovery behaviors in successful episodes. Project page: https://intuitive-robots.github.io/PHIL-page/.
Open source S4number of tasks
4
View original evidence
ce through an extended reality (XR) interface. PHIL pairs rejected policy predictions with corrective action chunks subsequently executed from the same physical state over the same horizon, and preference-based post-training encourages the policy to favor these corrections. This paper evaluates PHIL on four real-world manipulation tasks using a diffusion policy and a pretrained vision-language-action policy. With fiv
Open source S3
- paper: not_reported
- weights: not_reported
- dataset: not_reported
- code: not_reported
What remains unknown
Not established in the collected evidence: Environment, 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.
arXiv:2611.05667v1 Announce Type: new Abstract: This paper presents Proactive Human-in-the-Loop (PHIL), a framework for proactive policy correction and preference-based post-training. PHIL enables operators to preview policy-predicted future motion and intervene before anticipated undesirable actions. During takeover, PHIL continues querying the policy and displaying its predicted trajectories in the physical workspa
Open source S2
e correction episodes per condition, PHIL increases mean success rates across the evaluated tasks and demonstration budgets from 44.0% to 70.3% for the diffusion policy and from 57.8% to 76.9% for X-VLA, compared with HG-DAgger. Separately, policies trained with proactive corrections show fewer unnecessary recovery behaviors in successful episodes. Project page: https://intuitive-robots.github.io/PHIL-page/.
Open source S4
ce through an extended reality (XR) interface. PHIL pairs rejected policy predictions with corrective action chunks subsequently executed from the same physical state over the same horizon, and preference-based post-training encourages the policy to favor these corrections. This paper evaluates PHIL on four real-world manipulation tasks using a diffusion policy and a pretrained vision-language-action policy. With fiv
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org