Robot learning and a generalist–specialist debate on PEARL’s IROS 2026 agenda

In a September 27 announcement, a PEARL participant outlined plans for IROS 2026 in Pittsburgh. Research presentations are set to cover self-supervised multisensory pretraining for contact-rich robot reinforcement learning and scalable reinforcement-learning data generation for language-conditioned, two-handed dexterous manipulation. The author also plans to argue the generalist side in a generalist-versus-specialist plenary debate, though “probably not quite in the way the label suggests.” Invited workshop talks cover hybrid architectures for embodied autonomy, contact and learning, and geometric representations in robotics.