Cartender: Verifiable Language-Conditioned Driving by Learning the Drives Not Taken
Cartender: Verifiable Language-Conditioned Driving by Learning the Drives Not Taken
Cartender is a research paper proposing a method for language-conditioned driving that learns from alternative actions not taken in real-world data. It uses counterfactual branching to generate instruction-action pairs, improving instruction success and reward-based fidelity over traditional imitation learning approaches.
Cartender introduces a novel approach to language-conditioned driving by leveraging counterfactual branching to learn from unexecuted actions, improving instruction success and reward-based fidelity over traditional imitation methods.
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
RFS score
8
View original evidence
ity score regulates language conditioning in trajectory generation, and a reward-trained language residual re-ranks generated candidates over a scene-based score. The branches are used for training; instruction following is evaluated by swapping instructions under identical recorded inputs and measuring behavior rather than agreement with generated targets. On WOD-E2E, the best of four proposals achieves an RFS of 8.
Open source S4RFS score
8.203
View original evidence
642 across 3,603 feasible instruction conditions. Across 8,974 admitted proposals, mean RFS is 8.203 versus 7.906 for matched logged trajectories, and 67% score higher than their logged counterpart. Compared with matched-only imitation using the same backbone, Cartender improves same-scene paired instruction success from 23.67% to 57.00% and RFS from 6.32 to 7.10; its gate reduces infeasible-instruction execution fro
Open source S5RFS score
7.906
View original evidence
642 across 3,603 feasible instruction conditions. Across 8,974 admitted proposals, mean RFS is 8.203 versus 7.906 for matched logged trajectories, and 67% score higher than their logged counterpart. Compared with matched-only imitation using the same backbone, Cartender improves same-scene paired instruction success from 23.67% to 57.00% and RFS from 6.32 to 7.10; its gate reduces infeasible-instruction execution fro
Open source S5RFS score
6.32–7.1
View original evidence
642 across 3,603 feasible instruction conditions. Across 8,974 admitted proposals, mean RFS is 8.203 versus 7.906 for matched logged trajectories, and 67% score higher than their logged counterpart. Compared with matched-only imitation using the same backbone, Cartender improves same-scene paired instruction success from 23.67% to 57.00% and RFS from 6.32 to 7.10; its gate reduces infeasible-instruction execution fro
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ity score regulates language conditioning in trajectory generation, and a reward-trained language residual re-ranks generated candidates over a scene-based score. The branches are used for training; instruction following is evaluated by swapping instructions under identical recorded inputs and measuring behavior rather than agreement with generated targets. On WOD-E2E, the best of four proposals achieves an RFS of 8.
Open source S4
642 across 3,603 feasible instruction conditions. Across 8,974 admitted proposals, mean RFS is 8.203 versus 7.906 for matched logged trajectories, and 67% score higher than their logged counterpart. Compared with matched-only imitation using the same backbone, Cartender improves same-scene paired instruction success from 23.67% to 57.00% and RFS from 6.32 to 7.10; its gate reduces infeasible-instruction execution fro
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org