Launch Dexterity — Kevin Chavez Dexterity unveiled Foresight, its world model for robotic manipulation trained on over 100 million real warehouse actions, designed to let robots reason about physical change and predict outcomes before they happen. Introducing Foresight · Dexterity
Research Physical Intelligence — Charles Xu, Jost Tobias Springenberg, Michael Equi Physical Intelligence introduces RLT, a reinforcement-learning method that lets robots master precise manipulation tasks like screw alignment in hours rather than days of training. Precise Manipulation with Efficient Online RL · Physical Intelligence
Research Humanoid Humanoid's KinetIQ Ascend framework uses reinforcement learning to push its humanoid robots beyond imitation, achieving 42% higher throughput in machine-feeding tasks and 98% success rates in bin-picking. KinetIQ Ascend: Toward 100% Reliable Manipulation and Superhuman Speed · Humanoid
Research Physical Intelligence — Sergey Levine Physical Intelligence unveiled π 0.7, a robotic hand model that generalizes across dexterous tasks and responds to novel language commands without task-specific training. π 0.7 : a Steerable Model with Emergent Capabilities · Physical Intelligence