iDriveVLA is proposed, a multi-modal planning framework that improves the candidate trajectory space while enabling more reliable and context-aware trajectory evaluation and develops an oracle-aligned progressive training strategy consisting of candidate imitation pretraining, candidate space refinement, and semantic ranking alignment.
Abstract
Multi-modal planning is promising for autonomous driving by representing multiple plausible behaviors in ambiguous and long-tail scenarios. Existing methods mainly focus on improving trajectory multi-modality, enhancing trajectory representations, or reshaping the candidate distribution. Nevertheless, we identify a pronounced generation-evaluation asymmetry in multi-modal planning: despite strong oracle performance, existing planners often fail to reliably select the best available candidate, leaving substantial planning potential unrealized. To address this challenge, we propose iDriveVLA, a multi-modal planning framework that improves the candidate trajectory space while enabling more reliable and context-aware trajectory evaluation. Specifically, iDriveVLA introduces a unified trajectory evaluator comprising a Safety-aware Scorer for quality and risk estimation, together with a VLM-guided Modulator for scene-adaptive criterion weighting. We further develop an oracle-aligned progressive training strategy consisting of candidate imitation pretraining, candidate space refinement, and semantic ranking alignment. On the public NAVSIM v1 leaderboard, iDriveVLA achieves a new state-of-the-art performance of 94.95 PDMS, surpassing the human-expert reference.
Imitation learning planners can behave unsafely under distribution shift, since long-tailed driving data contain few safety-critical events. A common remedy is to generate multimodal candidate trajectories and select them with a hybrid evaluator, but existing methods often suffer from mode collapse and brittle hard con...
Wenxiao Ke, Wen-Jie Lu, Fu-Jun Peng et al.· IEEE Robotics and Automation...· 0 citations
This work designs a hybrid architecture that combines sparse anchors with an offset refinement module for efficient multi-modal trajectory prediction and adopts a two-stage training paradigm consisting of pretraining and reward-guided fine-tuning, demonstrating superior performance under real-time constraints.
Cheng-Lin Chen, Lu-Jia Wang, Xin-Hu Zheng et al.· 0 citations
MM-Future, a world-action model that generates multiple paired scene-action hypotheses and models bidirectional interaction within each pair, shows consistent improvements over both single-mode and action-only variants, validating the benefit of multi-mode joint world-action modeling.
Shuai Liu, Hechangle Gong, Hao Jiang et al.· 1 citation
S2Planner is presented, a trajectory planner that combines three front-facing cameras with ego-motion history and the current driving command, and a coarse-to-fine decoder then uses trajectory self-attention and camera-projected cross-attention to refine candidate waypoints.
Zhao-Wei Lu, Li-Guo Zhou, Yu-Jie Guo et al.· 0 citations
A novel framework that aligns multi-trajectory supervision with policy optimization, and introduces two complementary mechanisms: feasibility-first advantage assignment and dynamic distillation to ensure that expanded trajectory supervision is effectively absorbed during policy optimization.
Tian Zhang, Zhuo Huang, Hong-Rui Ye et al.· 0 citations
Autonomous driving requires choosing a safe and efficient plan as surrounding traffic evolves. Generate-and-select planners propose multiple trajectories and score them for execution, and they have outperformed representative direct-prediction baselines on NAVSIM. Their scorer must compare plans that were never execute...
Jie-Yuan Pei, Mei-Yi Lu, Sining Ang et al.· arXiv.org· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduJul 30, 2026
From feet to fingertips — we are teaching robots intelligent whole-body control, fine dexterity, and teamwork to complete a broad range of complex tasks.