$TF^2$-Steering: Action Refinement with Tactile Forecasts and Feedback via Velocity-Field Steering
$TF^2$-Steering: Action Refinement with Tactile Forecasts and Feedback via Velocity-Field Steering
This research introduces $TF^2$-Steering, a method that refines robot actions using tactile forecasts and feedback through velocity-field steering. By employing a tactile mixture-of-experts architecture, the approach corrects and redirects action generation based on tactile signals while leveraging vision-based foundation policies. Experiments across twelve tasks (six simulated, six real-world) show it outperforms baselines by 10.7 percentage points in simulation success rate and 9.1 percentage points in real-world progress success rate.
The study demonstrates that integrating tactile forecasts and feedback via velocity-field steering improves success rates in simulated and real-world manipulation tasks compared to vision-dominant baselines, though the results are based on experiments conducted by the authors.
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.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
of the vision-based foundation policy. Extensive experiments across six simulated and six real-world manipulation tasks demonstrate the effectiveness of $TF^2$-Steering. It outperforms the strongest evaluated baselines by 10.7 percentage points in average simulation success rate and 9.1 percentage points in average real-world progress success rate. Project page: https://yiming-cc.github.io/tf2-steering/.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org