AI
Alibaba Open Sourced a Self Driving Brain Anyone Can Inspect
Alibaba's Qwen team just released Qwen-Drive-1.0-4B, an open source model that reads a driving scene and plans what the vehicle should do next, and they published the whole thing. Code, model weights, and demo data all ship under the Apache 2.0 license, which means any researcher, startup, or curious engineer can download it, run it, and take it apart.
The model was developed with the Huazhong University of Science and Technology. It starts from Qwen3.5-4B as its vision language foundation, then adds dedicated components for 3D perception and for generating driving trajectories. The vision language core keeps its original architecture and its ability to answer visual questions, so the same brain that watches the road can also explain what it sees.
Two flavors of the planning component ship side by side. One learns by imitating driving examples. The other takes that imitation further with reinforcement learning. Publishing both lets the research community compare approaches openly, on identical foundations, which is how science is supposed to work.
Trajectory planning is the part of driving most people think about least and engineers think about most. Seeing the world is one challenge; deciding exactly where the car should be, second by second, is another. By open sourcing that layer alongside perception, the Qwen team handed the community the full loop, from pixels to path, ready to test, improve, and remix.
Open sourcing at this level also changes who gets to participate in the safety conversation. Regulators, academics, and independent researchers can probe the same model the developers ship, which builds public confidence faster than any marketing claim. When the weights are public, the audit trail is public too, and that transparency compounds across every team that forks the work. A student in Nairobi gets the same starting line as a lab in Silicon Valley, and the best ideas win on merit.
Four billion parameters keeps this firmly in the accessible zone. This is a model a university lab or a small robotics team can actually run and study, which matters because driving is safety critical AI. Closed robotaxi stacks ask for trust. Open weights invite verification, and verification is the stronger foundation for machines that share our streets.
What it means for you: open source self driving research just got a serious new building block, and progress compounds when everyone builds on the same foundation. Expect faster experimentation from labs everywhere, more eyes on safety, and a clearer path from research demo to real vehicles. The road to autonomy gets shorter when the map is public.
Quick answers
What is this story about?
Alibaba's Qwen team just released Qwen-Drive-1.0-4B, an open source model that reads a driving scene and plans what the vehicle should do next, and they published the whole thing. Code, model weights, and demo data all ship under the Apache 2.0 license, which means any researcher, startup, or curious engineer can download it, run it, and take it apart.
Why does this story matter?
What it means for you: open source self driving research just got a serious new building block, and progress compounds when everyone builds on the same foundation. Expect faster experimentation from labs everywhere, more eyes on safety, and a clearer path from research demo to real vehicles. The road to autonomy gets shorter when the map is public.
Sources
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.