MDrive: Benchmarking Closed-Loop Cooperative Driving for End-to-End Multi-agent Systems
Marco Coscoy, Zewei Zhou, Seth Z. Zhao, Henry Wei, Jiaqi Ma · · 2026
Framework
N/A
License
N/A
Stars
17
Summary
Vehicle-to-Everything (V2X) communication has emerged as a promising paradigm for autonomous driving, enabling connected agents to share complementary perception information and negotiate with each other to benefit the final planning.
Abstract Summary
Key Points
- Proposes MDrive, a novel approach for navigation in robotics.
- Addresses key limitations in existing methods through innovative architecture design.
- Demonstrates strong empirical results on standard benchmarks and real-world evaluations.
- Introduces novel training methodology that improves generalization and sample efficiency.
- Provides comprehensive analysis of failure modes and ablation studies.
MDrive Benchmark
MDrive is a closed-loop cooperative driving benchmark built on CARLA [ 55 ] simulator, which aims to provide broad, structured coverage of multi-agent scenarios to bridge the existing evaluation gap.
MDrive includes 225 scenarios and combines three complementary scenario buckets: 1) MDrive-Interdrive : in-domain performance of existing models without background actors; 2) MDrive-Interaction : interaction scenarios that enrich background actors interactivity and diversity; 3) MDrive-V2XPnP Real2Sim scenarios for testing models under real-world s...
Benchmarking State-of-the-Art Models
In this section, we present benchmark results to answer two research questions (RQ): RQ1: To what extent does multi-agent cooperation improve closed-loop driving over single-agent systems, and how do the gains decompose between perception sharing and negotiation? RQ2: What performance gaps and new challenges for multi-agent system do MDrive expose?
We benchmark on a set of representative single-agent and multi-agent end-to-end driving models.
Sources and Demonstrations
Figures are reproduced from Coscoy et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.
Related Papers
A Heterogeneous Architecture for Robot RL Beyond GPU-Dominant Paradigms
Yufei Jia, Zhanxiang Cao, Mingrui Yu et al.
Simulation-based RL for contemporary robot control is increasingly organized around GPU-resident simulation: physics, rollout collection, and learning are placed on a single GPU-centric execution path
EgoForce: Forearm-Guided Camera-Space 3D Hand Pose from a Monocular Egocentric Camera
Christen Millerdurai, Shaoxiang Wang, Yaxu Xie et al.
Reconstructing the absolute 3D pose and shape of the hands from the user's viewpoint using a single head-mounted camera is crucial for practical egocentric interaction in AR/VR, telepresence, and hand-centric manipulation tasks, where sensing must remain compact and unobtrusive.
Generalization of World Models under Environmental Variability for Vision-based Quadrotor Navigation
Luca Zanatta, Grzegorz Malczyk, Kostas Alexis
World models, learned generative models that predict how an environment evolves, have become a promising tool for sample-efficient robot learning. Yet how robust they are to environmental variability remains poorly understood.
GuideWalk: Learning Unified Autonomous Navigation and Locomotion for Humanoid Robots across Versatile Terrains
Haoxuan Han, Chen Chen, Linao Gong et al.
Humanoid robots have achieved strong locomotion capabilities, but reliable navigation on versatile terrains remains challenging because obstacle avoidance must be coordinated with dynamically feasible motion.