Alibaba’s Qwen team has released Qwen-Drive-1.0-4B, an open-source model that combines driving-scene understanding with planning a vehicle’s movements. Developed with Huazhong University of Science and Technology, it uses Qwen3.5-4B as its vision-language foundation and adds separate components for 3D perception and generating driving trajectories.
The release includes two versions of the planning component: one trained to imitate driving examples and another further optimized through reinforcement learning. Both use the same underlying vision-language model, which retains its original architecture and ability to answer visual questions. The project provides code, model weights and demo data under the Apache 2.0 license. [Qwen]