LAUSR

Abstract In this paper, the problem of robotic Sit-To-Stand (STS) assistance is studied. The objective is to effectively assist individuals in need to stand up from a seated position using a robot manipulator. To achieve the goal, we propose an integrated method which encompasses traditional model-based control and optimization, as well as AI-based human intention recognition. Specifically, a number of demonstrations of human-to-human STS assistance are first performed and recorded using motion capture system. On the account of the observation and recorded data, the average intended motion trajectories for the joints of lower limbs are obtained. Based on these intended motion trajectories as well as the constructed human body dynamics and control in different STS phases, an optimal nominal trajectory of the robot end-effector is generated off-line that minimizes the human joint loads while satisfying additional physical constraints. In actual STS assistance, the human who is being assisted is likely to move faster or slower from the nominal trajectories, or even sit back down. Therefore, we develop a Long Short-Term Memory (LSTM) network to estimate the ever-changing human’s intention in STS assistance, and then adjust the velocity of the robot end-effector on the basis of the predicted human intention on the nominal trajectory. Simulations and experiments are conducted, demonstrating that the proposed algorithm is indeed capable of minimizing joint load of human while following his/her intention during the course of STS motion. The algorithm can potentially be applied to future home robots that assist elderly and disabled people with daily activities.