Exploration of human control strategies for actuating elastic joints - an optimal control approach

Abstract

Elastic joints endow robots with advantages including energy-efficiency, safety and performance. To optimally actuate such joints, robotic tasks are formulated as optimal control problems with task objectives and their solutions provide the optimal control policies. However, two general challenges while solving these problems exist. They include the mathematical formulation of a task objective, due to a lack of verifiable control policies and the computation of the optimal solutions due to nonlinearities, state constraints, multiple degrees of freedom, etc. Since humans are considered to be optimizing agents with their innate constraints, a systematic analysis of their movement behaviour might provide clues to tackle these two issues. Therefore, we propose a novel approach where humans can be beneficially employed. For the same, we provide an analysis through experiments during which, subjects learned to manipulate virtual objects which represent elastic joints to perform two control tasks. In the first control task of maximizing the object’s terminal velocity, a cost functional can be easily formulated but the control policy computation is difficult. Humans while performing this task, showed that they are capable of approaching optimal task execution. Their apparent control strategies can be used as initial guesses to reduce the computational difficulties in finding optimal solutions. In the second control task of displacing an elastic object in the shortest time, finding a criterion that can represent a cost function is not straightforward. The cost functionals that were optimized by humans when deciphered, provided insights about the relative weighting of the dominant costs including hand acceleration, hand and object distance. These findings favour the exploration and utilization of human control policies for actuating elastic joints in future research.