Nonlinear Optimization for Human-Like Movements of a High Degree of Freedom Robotics Arm-Hand System

Abstract

The design of autonomous robots, able to closely cooperate with human users in shared tasks, provides many new challenges for robotics research. Compared to industrial applications, robots working in human environments will need to have human-like abilities in their cognitive and motor behaviors. Here we present a model for generating trajectories of a high degree of freedom robotics arm-hand system that reflects optimality principles of human motor control. The process of finding a human-like trajectory among all possible solutions is formalized as a large-scale nonlinear optimization problem. We compare numerically three existing solvers, IPOPT, KNITRO and SNOPT, in terms of their real-time performance in different reach-to-grasp problems that are part of a human-robot interaction task. The results show that the SQP methods obtain better results than the IP methods. SNOPT finds optimal solutions for all tested problems in competitive computational times, thus being the one that best serves our purpose.

Generating human-like movements on an anthropomorphic robot using an interior point method

“…For the sake of brevity, some details of the model and the kinematics of the robotic system are omitted (for further details see [1]). The anthropomorphic redundant robotic arm and hand ( Figure 1) can be represented as a series of links connected by joints.…”

“…Here we simplified Pa and Pb as presented in [1]. First, since the middle finger is opposite to the other two we set θ f ,8 = 0.…”

“…where θ m and θ M are constant vectors that represent the lower and upper joint limits, δ > 0 is a constant, c 1 and c 2 are nonlinear functions (of target pose (position and orientation) and joint angles, obtained using direct kinematics) concerning the position and orientation of the robot hand relatively to the target, respectively, ε(t i ) is a function of time representing the clearance distance, h f , h b , h b are nonlinear functions of the obstacles pose and of the arm and/or finger angles. For a more extensive description of the constraints see [1].…”

“…For achieving natural and efficient human-robot interaction and collaboration, human-like movements are essential, since it allows the human involved in the interaction to interpret the robot's movements. Inspired by the Posture-Based Motion-Planning Model (PBMP) of human movements [2], we have, in previous work [1], presented a model for generating human-like arm and hand movements on an anthropomorphic robot involved in human-robot collaboration tasks (see [3]). The previous results focused on reach-to-grasp movements with two different grip types [1].…”

“…Inspired by the Posture-Based Motion-Planning Model (PBMP) of human movements [2], we have, in previous work [1], presented a model for generating human-like arm and hand movements on an anthropomorphic robot involved in human-robot collaboration tasks (see [3]). The previous results focused on reach-to-grasp movements with two different grip types [1]. Here we extend our model in order to address the generation of more complex movement sequences (see Figure 1) which are challenged by scenarios cluttered with obstacles.…”