Autoware 2.0 lets AI and rule-based planners run inside one driving stack

The Autoware Foundation released Autoware 2.0 on November 2, adding support for end-to-end AI and hybrid driving stacks alongside the rule-based modules the open-source autonomous driving platform already carried. The Foundation said the release builds on the existing Autoware Core and Autoware Universe layers rather than replacing them.

The central change is a generator–selector architecture, in which several trajectory generators run inside the same system and a selector decides which output is used. Developers can compose one architecture from rule-based modules, AI modules or a combination, and the Foundation shipped documentation and a simulator demonstration with it. Model handling moved too: its Hugging Face organization became the versioned default home for Foundation models and datasets, replacing what it said had been a single central Ansible file defining every model download.

The release also introduced the Autoware Index, a registry of compatible repositories with a command-line tool that writes a chosen set of packages into a workspace file, and the Autoware Reference Design framework. A Reference Design pairs a Deployment Profile — vehicle and hardware configuration, operating conditions and use case — with a maintained Reference Software Stack, describes its operating domain in ASAM OpenODD terminology, and states its status so that intended capability is distinguished from demonstrated results. Ryohsuke Mitsudome, the Foundation’s technical steering committee chair, whom its organization page lists under TIER IV, said users often ask how to get from Autoware’s software to a complete autonomous vehicle.

TIER IV said separately the same day that it would lead the 2.0 initiatives, naming work with Arm on computing platforms, with AWS on cloud-native development and operations, and with Red Hat on integrating 2.0 with AutoSD, the open-source upstream of Red Hat In-Vehicle Operating System. Founder and chief executive Shinpei Kato said open-source software alone was not enough to transform the automotive industry. The role descriptions are TIER IV’s: none of the three had announced anything in its own newsroom when checked. The GitHub release tag for 2.0.0 was published on October 29, four days before the announcement.