TIER IV, the company behind the open-source Autoware self-driving stack, released two new autonomous driving AI models on November 11, 2026: Alpamayo 2 Edge, developed with Nvidia, and OnePlanner, built in-house. Both are now integrated with Autoware and published openly on GitHub and Hugging Face. Alpamayo 2 Edge is a camera-only, end-to-end model that takes in images and uses reasoning to generate driving trajectories directly, compressed down to 2 billion parameters from the 34 billion in its sibling, Alpamayo 2 Super. OnePlanner, meanwhile, fuses perception and motion planning into a single step, using BEVFusion architecture to generate six-second driving trajectories that incorporate HD maps, route data, traffic signals and vehicle motion states.
Why Shrinking the Model Matters More Than Growing It
Most of the industry's attention on autonomous driving AI has gone toward larger models with more sensors — lidar, radar, multiple cameras — stacked for redundancy. Alpamayo 2 Edge runs the opposite direction: cutting parameter count by 94% while keeping the model camera-only. That matters because lidar and high-end compute remain the two biggest cost centers in building a self-driving vehicle, and a model light enough to run on modest edge hardware opens the door to lower-cost platforms such as delivery vehicles, shuttles and retrofit kits for existing fleets, rather than purpose-built robotaxis with six-figure sensor stacks.
OnePlanner addresses a different inefficiency: current pipelines typically convert perception data into an intermediate format before a separate planning module consumes it, losing information at each handoff. By integrating perception directly into planning, TIER IV is betting that fewer conversion steps mean richer use of the same sensor data, without necessarily buying more of it.
An Open Approach Against Closed Competitors
TIER IV's strategy is distinct from most of the autonomous driving field. Waymo, Zoox and Tesla all keep their driving stacks proprietary, treating the model as the core competitive moat. TIER IV instead publishes Autoware and now these two models under open licenses, betting on adoption breadth rather than exclusivity — a strategy closer to what Nvidia itself has pursued with other open robotics and driving tools, and similar in spirit to Runway's recent open release of its Praxis-1 world action model for robot arms.
The trade-off is that open models invite scrutiny and forking faster than they generate direct revenue. TIER IV's business model leans on selling integration services, hardware partnerships and consulting around Autoware rather than licensing the model itself, a path that depends on becoming the default toolkit before a closed competitor locks in a given vehicle segment. With Alpamayo 2 Edge still in the optimization phase for real vehicles rather than production-ready, the near-term test is whether outside developers actually adopt it at meaningful scale, or whether it remains a research artifact alongside dozens of other open driving models racing for attention.
Source: PR Newswire via Manila Times