Driving · 2026-10-14
Drive3R: Block-wise Streaming Geometric Foundation Models for Autonomous Driving
Tianyi Yang, Haojun Yang, Jipeng Xiao, Xuesong Chen, Shaoshuai Shi, Li Jiang, Zhipeng Zhang
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
Drive3R is a block-wise streaming geometric foundation model for long-horizon autonomous driving reconstruction. It uses block-wise causal attention to keep bidirectional context within each block while enforcing causality across blocks via KV-cache reuse. It also adds confidence-aware re-anchoring that discards stale KV cache and stitches refreshed segments using SE(3) alignment.
Why it matters
Applying geometric foundation models to continuous driving sequences is computationally intractable with full bidirectional attention (quadratic attention and GPU memory). The authors aim to preserve local bidirectional reasoning while enabling scalable streaming and reducing long-range drift via adaptive re-anchoring.
Method
- Block-wise causal attention: tokens within the same temporal block are mutually visible; future blocks are masked, while streaming inference reuses historical KV cache across blocks.
- Streaming with KV cache: at inference, each block attends to current-block keys/values plus cached keys/values from previous blocks; after processing, its keys/values are appended to the cache.
- Confidence-aware re-anchoring: a block-level confidence score triggers cache reset and SE(3) alignment stitching on overlapping frames when drift is detected.
Abstract (from arXiv)
High-fidelity closed-loop simulation for autonomous driving increasingly relies on dense multi-view 3D reconstruction. While recent feed-forward geometric foundation models have revolutionized per-scene optimization, applying them to continuous, long-horizon driving sequences is computationally intractable due to GPU memory limits and the quadratic complexity of global attention. Existing scaling strategies are fundamentally limited, as causal streaming suffers from compounding scale drift and trajectory deviation, while rigid "chunk-then-align" heuristics fragment global context and produce trajectory discontinuities. To address these limitations, we introduce Drive3R, a block-wise streaming geometric foundation model. Its block-wise causal attention resolves the scalability bottleneck of fully bidirectional attention while avoiding the loss of local future context in purely causal streaming. Its confidence-aware re-anchoring further mitigates long-range drift by detecting unreliable streaming states, re-initializing the KV cache, and stitching refreshed segments through SE(3) alignment. Extensive experiments demonstrate that Drive3R establishes a new state-of-the-art on nuScenes, KITTI, and Waymo. Specifically, on nuScenes, it improves over the recent DVGT-2 baseline from 0.1806 to 0.1081 in AbsRel, from 0.6750 to 0.5344 in point-cloud Accuracy, and from 1.5812 to 0.9092 in ATE.
Related papers
- DriveWorld3D Evidence Constrained 3D World Modeling and Generative Simulation for Real-World Autonomous Driving
- GeoV2V: Geometry-Grounded Video Diffusion Model for Driving Scene Generation
- DriveDreamer: Towards Real-world-driven World Models for Autonomous Driving
- PhysWAM: Physically Consistent World Action Model for Autonomous Driving
- TerraTransfer++: Demonstration-Free End-to-End Planner
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-10 · Indexed by Pub-AI: 2026-10-14 UTC
Drive3R: Block-wise Streaming Geometric Foundation Models for Autonomous Driving
Preprint posted: 2026-10-10 · Original record ↗
- arxiv
- 2610.15112
Explicit source versions (1)
- v1 · 2026-10-10
- Preprint available
Source date: 2026-10-10 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
High-fidelity closed-loop simulation for autonomous driving increasingly relies on dense multi-view 3D reconstruction. While recent feed-forward geometric foundation models have revolutionized per-scene optimization, applying them to continuous, long-horizon driving sequences is computationally intractable due to GPU memory limits and the quadratic complexity of global attention. Existing scaling strategies are fundamentally limited, as causal streaming suffers from compounding scale drift and trajectory deviation, while rigid "chunk-then-align" heuristics fragment global context and produce trajectory discontinuities. To address these limitations, we introduce Drive3R, a block-wise streaming geometric foundation model. Its block-wise causal attention resolves the scalability bottleneck of fully bidirectional attention while avoiding the loss of local future context in purely causal streaming. Its confidence-aware re-anchoring further mitigates long-range drift by detecting unreliable streaming states, re-initializing the KV cache, and stitching refreshed segments through SE(3) alignment. Extensive experiments demonstrate that Drive3R establishes a new state-of-the-art on nuScenes, KITTI, and Waymo. Specifically, on nuScenes, it improves over the recent DVGT-2 baseline from 0.1806 to 0.1081 in AbsRel, from 0.6750 to 0.5344 in point-cloud Accuracy, and from 1.5812 to 0.9092 in ATE.