Tier IV is betting that end-to-end autonomous driving will spread faster if automakers can choose their own models and chips
Tier IV has released open-source end-to-end autonomous driving AI models through Autoware, in configurations from camera-only setups to systems combining LiDAR, radar and high-definition maps. Developers can pick a model to suit their performance targets, with Autoware supplying the tools to adapt and integrate it into vehicles.
The models are built for Autoware 2.0, a Tier IV-led architecture pairing AI with rule-based software. Their breadth reflects Tier IV’s view that automakers and suppliers should be able to adopt AI without relying on one model or chip platform.
Nvidia’s Alpamayo 2 Edge, a camera-only reasoning model integrated with Autoware, has two billion parameters against Alpamayo 2 Super’s 34 billion. AutoE2E, co-developed with the Autoware Foundation, runs beyond Nvidia hardware at 79.9 million parameters and is being assessed for robotaxis.
Tier IV’s own Meteor is lighter still at 16 million parameters, targeting private cars on Nvidia and Renesas chips without high-definition maps. OnePlanner and LatentDrive complete the line-up, planning six and four seconds ahead from LiDAR and camera inputs respectively.
In a statement, Ali Kani, Vice President of Automotive at Nvidia, said: “Advancing autonomous driving requires AI that can reason through complex situations and run efficiently in the car. Tier IV’s integration of Nvidia Alpamayo 2 Edge with Autoware gives developers an open foundation to bring that intelligence from research into their own vehicles.”
Why this matters:
- Open-source challenges proprietary E2E stacks. Publishing end-to-end models alongside an integration environment could let smaller automakers and suppliers build Level 4 systems without licensing a full proprietary stack, though validation and liability would still rest with them.
- Hardware independence as strategic hedge. Offering AutoE2E for non-Nvidia platforms suggests developers want end-to-end autonomy to remain portable across compute suppliers, as chip choices become increasingly entangled with cost pressures and geopolitics.
- Rule-based layers ease certification. Combining AI models with conventional rule-based technology suggests hybrid architectures may offer the more practical route to approval while regulators remain cautious about certifying purely learned driving behaviour.
Source: www.automotiveworld.com
