Driving · 2026-11-13

CARE-Drive: Coupled Alternating Refinement for Ego-trajectories

Xiao Ma, Yijun Chen, Zhiquan Hu, Jicheng Zhao, Yuming Li, Chuang Dai

arXiv:2611.15360PDFCode

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

CARE-Drive (Coupled Alternating Refinement for Ego-trajectories) refines ego trajectories by separating waypoint residual correction from position-conditioned heading generation. A future-token auxiliary task trains a predictive VLA backbone; then, with the backbone frozen, two specialists are optimized alternately (supervised initialization + reinforcement fine-tuning). On NAVSIM v1, the authors report 88.8 PDMS, +2.4 over the unrefined backbone.

Why it matters

The authors argue that unified trajectory refinement underemphasizes distinct physical quantities and their interactions. CARE-Drive explicitly specializes refinement into position and heading experts, with heading conditioned on corrected positions, and trains them alternately while keeping the backbone frozen. The reported NAVSIM v1 gains quantify how this structured, coupled refinement improves planning quality versus the same unrefined backbone.

Method

  • Backbone trained with trajectory imitation plus a future-token auxiliary task: future query tokens predict latent future image features from a frozen video teacher (SmoothL1 loss).
  • Position expert (frozen backbone features): autoregressive transformer predicts bounded metric waypoint residuals; heading expert reconstructs orientation from adjacent angular increments conditioned on corrected XY.
  • Alternating optimization with backbone frozen: staged supervised initialization followed by reinforcement fine-tuning using PDMS-based reward/advantages (clipped GRPO with KL regularization).

Limitation

The paper does not state any explicit limitation in the provided text.

Abstract (from arXiv)

Vision-language-action (VLA) models bring multimodal understanding and world knowledge to end-to-end autonomous driving. To further improve the quality of predicted trajectories, existing studies introduce candidate refinement, residual denoising, and token correction. However, these methods typically focus on unified refinement of the entire trajectory, paying limited attention to the distinct physical quantities and their interactions. We further investigate physical specialization and coordination in trajectory refinement, centering on position and heading for simpler and more intuitive correction modeling. To this end, we propose CARE-Drive (Coupled Alternating Refinement for Ego-trajectories), which decomposes trajectory refinement into waypoint residual correction and position-conditioned heading generation. Through imitation learning and a future-token auxiliary task, the backbone provides predictive representations and a base trajectory. The position expert predicts bounded metric waypoint residuals, while the heading expert reconstructs vehicle orientation from adjacent angular increments conditioned on the corrected path. With the backbone frozen, the two experts are optimized alternately through supervised initialization and reinforcement fine-tuning. This design simplifies refinement through clear physical specialization while preserving conditional coupling during prediction and optimization. On NAVSIM v1, CARE-Drive achieves 88.8 PDMS, improving by 2.4 points over the unrefined base trajectories generated by the same backbone. Ablation studies further support the effectiveness of specialized experts and staged training.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First announced by arXiv: 2026-11-13 · Indexed by Pub-AI: 2026-11-13 UTC

preprint

CARE-Drive: Coupled Alternating Refinement for Ego-trajectories

Xiao Ma, Yijun Chen, Zhiquan Hu, Jicheng Zhao, Yuming Li, Chuang Dai

First announced by arXiv: 2026-11-13 · Original record ↗

arxiv
2611.15360
Explicit source versions (1)
  • v1 · Date not supplied
  1. Preprint available

    Source date: 2026-11-13 · First detected: 2026-11-13 UTC · Historical baseline

    Source evidence (1)
    • Official evidence ↗ · RSS guid=oai:arXiv.org:2611.15360v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv

Vision-language-action (VLA) models bring multimodal understanding and world knowledge to end-to-end autonomous driving. To further improve the quality of predicted trajectories, existing studies introduce candidate refinement, residual denoising, and token correction. However, these methods typically focus on unified refinement of the entire trajectory, paying limited attention to the distinct physical quantities and their interactions. We further investigate physical specialization and coordination in trajectory refinement, centering on position and heading for simpler and more intuitive correction modeling. To this end, we propose CARE-Drive (Coupled Alternating Refinement for Ego-trajectories), which decomposes trajectory refinement into waypoint residual correction and position-conditioned heading generation. Through imitation learning and a future-token auxiliary task, the backbone provides predictive representations and a base trajectory. The position expert predicts bounded metric waypoint residuals, while the heading expert reconstructs vehicle orientation from adjacent angular increments conditioned on the corrected path. With the backbone frozen, the two experts are optimized alternately through supervised initialization and reinforcement fine-tuning. This design simplifies refinement through clear physical specialization while preserving conditional coupling during prediction and optimization. On NAVSIM v1, CARE-Drive achieves 88.8 PDMS, improving by 2.4 points over the unrefined base trajectories generated by the same backbone. Ablation studies further support the effectiveness of specialized experts and staged training.