Researchers at the University of Bristol and Bristol Robotics Laboratory have developed an innovative bi-touch system that allows robots to perform delicate bimanual tasks. The system leverages AI and tactile sensing to enable robot dexterity rivaling human precision.

The Bristol team created a tactile dual-arm robot that utilizes Deep Reinforcement Learning. Through this training method, the robot learns bimanual skills by attempting tasks and receiving virtual "rewards" or "punishments." This trial-and-error process mimics how humans and animals acquire new abilities.

The researchers first designed a virtual environment with two simulated robot arms equipped with tactile sensors. They then programmed reward functions and goal updates to encourage the AI agent to complete bimanual assignments. Finally, they applied the trained agent to a real-world tactile dual-arm robot.

Lead author Yijiong Lin explained, “With our Bi-Touch system, we can easily train AI agents in a virtual world to achieve bimanual tasks. More importantly, we can directly transfer these agents to the real world without retraining.”

The tactile AI agent can adapt to perturbations and gently manipulate fragile items. For example, it learned to lift objects like a single potato chip without dropping or breaking them. The robot performs these feats without visual data, relying solely on tactile and proprioceptive feedback.

Co-author Professor Nathan Lepora noted, “Our Bi-Touch system demonstrates an affordable approach to learning bimanual behaviors with touch in simulation, which directly applies to reality.”

The Bristol team's tactile robot simulation will be open-source to enable further research. This pioneering work could expand robot capabilities for delicate tasks like fruit picking, domestic service, and providing touch feedback in prosthetic limbs.

By mimicking human tactile learning, the University of Bristol's dual-arm robot showcases a promising new paradigm for dexterous robotics. The Bi-Touch system's combination of AI, simulation and tactile sensing could profoundly impact future automation technologies.