Humanoid Whole-Body Movement Optimization from Retargeted Human Motions

Abstract

Motion retargeting and teleoperation are powerful tools to demonstrate complex whole-body movements to humanoid robots: in a sense, they are the equivalent of kinesthetic teaching for manipulators. However, retargeted motions may not be optimal for the robot: because of different kinematics and dynamics, there could be other robot trajectories that perform the same task more efficiently, for example with less power consumption. We propose to use the retargeted trajectories to bootstrap a learning process aimed at optimizing the whole-body trajectories w.r.t. a specified cost function. To ensure that the optimized motions are safe, i.e., they do not violate system constraints, we use constrained optimization algorithms. We compare both global and local optimization approaches, since the optimized robot solution may not be close to the demonstrated one. We evaluate our framework with the humanoid robot iCub on an object lifting scenario, initially demonstrated by a human operator wearing a motion-tracking suit. By optimizing the initial retargeted movements, we can improve robot performance by over 40%.

Multi-Objective Trajectory Optimization to Improve Ergonomics in Human Motion

“…Similarly to [26], all ProMP trajectories can be stacked into a single weight vector, that finally defines our parameters to be optimized:…”

“…Single-Objective Trajectory Optimization (SOTO): We bootstrap the optimization with the initial ProMP weights learned from the demonstration set. To optimize each one of the scores separately, we use single-objective optimization with the optimizer COBYLA (Constrained Optimization BY Linear Approximation) [27], a deterministic local optimizer that directly takes black-box constraints as inputs alongside any of the ergonomics scores accumulated by (1), and has already been used for constrained motion optimization problems [26]. The COBYLA implementation is taken from the C++ library NLopt [28].…”

Automatic Tuning and Selection of Whole-Body Controllers

“…One well-known and often used evolutionary method is CMA-ES [10]. The authors of [11], [12] and [13] employ CMA-ES to search for optimal trajectory parameters in the form of basis function weights for simulated tasks on the child-sized humanoid iCub [14]. The approach of [12] uses a two-step optimization, the first of which is unconstrained CMA-ES to find a feasible starting point for the second step.…”

“…The approach of [12] uses a two-step optimization, the first of which is unconstrained CMA-ES to find a feasible starting point for the second step. On the other hand, authors of [11] initialize with hand-tuned parameters, while those of [13] skip the bootstrapping step by using retargeted recorded human trajectories to initialize the optimization problem. Both of these studies highlight an important weakness of CMA-ES: it must be initialized within the feasible region to find a solution.…”

Passivity-based control for haptic teleoperation of a legged manipulator in presence of time-delays

“…Passive whole-body controllers have been proposed to compliantly and safely interact with the environment [12], [13]. However, in a haptic teleoperation scenario, existing work has focused on bounding [14] and mapping [15] the operator commands to ensure safe teleoperation and maintain balance, rather than designing a controller that is inherently robust against the effects of time-delays. One way of addressing this problem is through an energy dissipation performed at the output of the whole-body controller, however this could interfere with the constrained optimal solution and render it infeasible.…”