Carlo D'Eramo

A frozen graph backbone effectively preserves past structural knowledge in continual multi-agent reinforcement learning, dramatically reducing forgetting and boosting transfer across tasks with distinct graph structures.

SPDL not only accelerates learning for autonomous superbike racing but also establishes a crucial baseline for future RL applications in complex two-wheeled dynamics.

Automating the RL workflow with HARBOR cuts engineering costs and enhances policy transferability to real-world robots.