Local LLMs go multimodal, Vercel Functions expand

Data Point

V2V-VLA model unifies cooperative autonomous driving

CMU-Drive benchmark and V2V-VLA model enable end-to-end cooperative driving by generating actions, waypoints, reasoning, and communication policies in a single forward pass.

Developers building multi-agent autonomous systems now have a closed-loop benchmark to evaluate cooperative perception and planning. This shifts from single-agent VLA models to explicitly testing vehicle-to-vehicle coordination under safety-critical conditions.

Replaces single-agent VLA baselines with a multi-agent cooperative framework. Requires integration with the CMU-Drive benchmark and V2V-VLA architecture. Early-stage research release—useful for benchmarking cooperative AV approaches, but production viability depends on real-world scenario coverage not yet detailed.

• “Vision-Language-Action (VLA) models have recently achieved impressive performance for end-to-end autonomous driving”
• “a closed-loop end-to-end benchmark for evaluating cooperative autonomous driving with multiple connected autonomous vehicles (CAVs)”
• “jointly generating driving actions, future waypoints, language reasoning, and communication policies”
• “Our code, benchmark, and model checkpoint will be publicly released”