DART2: Distributed AI for Robust and Trustworthy Autonomous Vehicles (phase 2)

Purpose and goal

DART2 develops distributed and privacy-preserving AI methods for autonomous vehicles through a collaboration between RISE, Zenseact, and NVIDIA FLARE. The project investigates how federated learning can be combined with self-supervised learning and knowledge distillation to enable robust and trustworthy AI training across distributed vehicle fleets without sharing raw data. The results will strengthen Sweden–US collaboration and support trustworthy AI solutions for safety-critical systems.

Expected effects and result

The project will evaluate NVIDIA FLARE for distributed AI in autonomous-vehicle applications and validate self-supervised and knowledge-distillation approaches in cross-silo settings. Expected results include joint scientific publications, a long-term Sweden–US research agenda, and technical insights enabling scalable and privacy-preserving AI. The outcomes will strengthen Sweden’s position in trustworthy AI and support future industrial deployment in autonomous systems.

Planned approach and implementation

The project combines industrial use cases from Zenseact with distributed AI expertise from RISE and NVIDIA FLARE. Through joint experiments, on-site discussions in Europe and the United States, and collaborative evaluation, the partners will investigate federated learning, self-supervised, and knowledge-distillation approaches using autonomous-driving data. The work is organized into three work packages covering use cases, technical experimentation, dissemination and long-term collaboration.