Who Taught the Robot That? Removing One User's Corrections from a Robot Policy That Learns Online
Who Taught the Robot That? Removing One User's Corrections from a Robot Policy That Learns Online
This research investigates the ability of robot policies trained online to forget specific human corrections. The authors show that if users correct the robot by taking over control, retraining on the remaining sessions in their original order allows the robot to approximate the policy as if the removed user never existed. However, simply deleting data or when users only nudge actions fails to fully remove the user's influence due to dependencies in the training buffer. The study concludes that while logs can identify harmful users, ranking them by influence is limited.
The study demonstrates that removing a specific user's corrections from an online-learning robot policy is feasible only when users take over control, as retraining on the remaining ordered log of sessions recovers the target policy; however, removal fails when users merely nudge actions because residual influence persists in the training buffer.
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.
- Environment
- SimulationOpen source S2
- Control
- ScriptedOpen source S2
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2611.10280v1 Announce Type: new Abstract: A robot that keeps learning after deployment is corrected by several people in turn, and one of them may later ask to be removed. We show that whether the robot can forget that person depends on how people correct it. The goal of removal is the robot that would exist if the person had never been there, which we compute exactly in two simulated tasks with scripted users.
Open source S2
Source：arXiv Robotics — research abstracts · arxiv.org