Robot policy gains now depend on memory, targeted adaptation, and harder dexterity tests
Robot learning work in this period concentrates on making existing policies more dependable under perturbations and sparse feedback. Harness VLA adds planning and retries around a frozen controller.