Blending of brain-machine interface and vision-guided autonomous robotics improves neuroprosthetic arm performance during grasping.

BACKGROUND: Recent studies have shown that brain-machine interfaces (BMIs) offer great potential for restoring upper limb function. However, grasping objects is a complicated task and the signals extracted from the brain may not always be capable of driving these movements reliably. Vision-guided robotic assistance is one possible way to improve BMI performance. We describe a method of shared control where the user controls a prosthetic arm using a BMI and receives assistance with positioning the hand when it approaches an object. METHODS: Two human subjects with tetraplegia used a robotic arm to complete object transport tasks with and without shared control. The shared control system was designed to provide a balance between BMI-derived intention and computer assistance. An autonomous robotic grasping system identified and tracked objects and defined stable grasp positions for these objects. The system identified when the user intended to interact with an object based on the BMI-controlled movements of the robotic arm. Using shared control, BMI controlled movements and autonomous grasping commands were blended to ensure secure grasps. RESULTS: Both subjects were more successful on object transfer tasks when using shared control compared to BMI control alone. Movements made using shared control were more accurate, more efficient, and less difficult. One participant attempted a task with multiple objects and successfully lifted one of two closely spaced objects in 92 % of trials, demonstrating the potential for users to accurately execute their intention while using shared control. CONCLUSIONS: Integration of BMI control with vision-guided robotic assistance led to improved performance on object transfer tasks. Providing assistance while maintaining generalizability will make BMI systems more attractive to potential users. TRIAL REGISTRATION: NCT01364480 and NCT01894802 .

Learning strategies in table tennis using inverse reinforcement learning.

Learning a complex task such as table tennis is a challenging problem for both robots and humans. Even after acquiring the necessary motor skills, a strategy is needed to choose where and how to return the ball to the opponent's court in order to win the game. The data-driven identification of basic strategies in interactive tasks, such as table tennis, is a largely unexplored problem. In this paper, we suggest a computational model for representing and inferring strategies, based on a Markov decision problem, where the reward function models the goal of the task as well as the strategic information. We show how this reward function can be discovered from demonstrations of table tennis matches using model-free inverse reinforcement learning. The resulting framework allows to identify basic elements on which the selection of striking movements is based. We tested our approach on data collected from players with different playing styles and under different playing conditions. The estimated reward function was able to capture expert-specific strategic information that sufficed to distinguish the expert among players with different skill levels as well as different playing styles.