TIER IV integrates Autoware with R-Car Gen 5

TIER IV and Renesas Electronics are combining the Autoware autonomous-driving software stack and reference AI models with Renesas R-Car Gen 5 automotive processors to create an open computing platform for software-defined vehicles. The collaboration is intended to provide a common development route from Level 2+/2++ driver assistance through Level 4 autonomous-driving systems.

TIER IV will adopt R-Car Gen 5 as one of its principal reference computing platforms and port and optimise Autoware for the processor family. Its end-to-end reference AI models are also due to run on the hardware, giving vehicle manufacturers and Tier 1 suppliers a defined combination of automotive SoC, open-source driving software, and model implementation rather than leaving each organisation to establish the complete hardware-software baseline independently.

The first public demonstration is planned around the R-Car X5H, Renesas’ high-end Gen 5 device for centralised automotive computing. R-Car Gen 5 is being positioned around a scalable architecture spanning higher-performance application processing and real-time vehicle control, with Renesas attempting to preserve software reuse as manufacturers move functions away from numerous discrete ECUs towards more centralised or zonal electrical architectures.

That transition is one reason software-defined vehicle programmes are demanding more from the underlying processor. Driver assistance and autonomous systems have to ingest multiple camera, radar, LiDAR, positioning, and vehicle-state inputs while running perception, prediction, planning, control, and increasingly AI-heavy processing within defined latency, thermal, and power limits.

End-to-end AI models alter that balance further. Conventional autonomous-driving stacks divide many tasks into separately engineered modules, making interfaces between perception, prediction, planning, and control relatively explicit. End-to-end models can learn larger sections of that chain directly, potentially simplifying some hand-crafted logic while increasing dependence on accelerator throughput, model training, validation data, determinism, and the ability to explain how a particular result emerged.

Porting those models onto production-oriented automotive silicon therefore involves more than proving that an inference network runs. Memory bandwidth, data formats, accelerator scheduling, sensor timing, real-time CPU workloads, operating systems, safety partitions, and thermal limits have to be considered together. A model developed on data-centre hardware may require substantial restructuring or quantisation before it can operate within a vehicle’s electrical and cooling envelope.

Autoware gives TIER IV and Renesas an existing open-source software foundation around which that integration can take place. A reference platform can reduce duplicated bring-up work by giving engineering teams a known processor, software stack, and AI implementation to evaluate before adapting the architecture to their own sensors, vehicle networks, and functional requirements.

Open does not mean interchangeable without qualification. Automotive programmes still have to address functional safety, cybersecurity, sensor failure, software update mechanisms, deterministic execution, long-term component availability, and the evidence required to release automated-driving functions onto public roads. An OEM adopting the reference stack remains responsible for the completed vehicle system and the behaviour created by its own configuration and training data.

The collaboration is nevertheless useful because the boundary between semiconductor selection and software architecture is becoming harder to maintain. Processor choices made early in a vehicle programme can determine which AI models remain practical years later, while software decisions can establish memory and accelerator requirements before the final silicon configuration is frozen.

TIER IV is also developing technology around future AI accelerators, LiDAR, and 4D radar, so the reference platform will have to accommodate changing sensing and processing requirements rather than represent a fixed end state. Renesas faces the corresponding challenge of keeping its SoC roadmap sufficiently stable for automotive qualification while giving software developers access to increasing compute and accelerator performance.

A demonstration of Autoware and the reference AI models on R-Car X5H is expected at Automotive World in September. That will establish whether the current software has progressed from architectural collaboration to working hardware, but the more important tests will follow in vehicle integration, safety validation, thermal performance, and production programmes. The partnership’s value will ultimately depend on whether a common reference platform removes enough engineering work to shorten a real vehicle development cycle.