CHOMP: Covariant Hamiltonian optimization for motion planning

Abstract

In this paper, we present CHOMP (Covariant Hamiltonian Optimization for Motion Planning), a method for trajectory optimization invariant to reparametrization. CHOMP uses functional gradient techniques to iteratively improve the quality of an initial trajectory, optimizing a functional that trades off between a smoothness and an obstacle avoidance component. CHOMP can be used to locally optimize feasible trajectories, as well as to solve motion planning queries, converging to lowcost trajectories even when initialized with infeasible ones. It uses Hamiltonian Monte Carlo to alleviate the problem of convergence to high-cost local minima (and for probabilistic completeness), and is capable of respecting hard constraints along the trajectory. We present extensive experiments with CHOMP on manipulation and locomotion tasks, using 7-DOF manipulators and a rough-terrain quadruped robot.

Effects of Robot Motion on Human-Robot Collaboration

“…Participants were surprisingly willing to move at the same time as the robot, and mainly complained about not being able to coordinate. Furthermore, functional motion does not require optimization, making it at times faster at producing a feasible plan [21].…”

Motion planning with sequential convex optimization and convex collision checking

“…In this section, we compare our approach vis-à-vis CHOMP (Ratliff et al, 2009;Zucker et al, 2012) and sampling-based motion planners (LaValle, 2006), and discuss the importance of trajectory initialization for trajectory optimization methods.…”

“…While the motivation for the presented work is very similar to the motivation behind CHOMP (Ratliff et al, 2009;Dragan et al, 2011;Zucker et al, 2012), which is most similar in terms of prior art, our algorithm differs fundamentally in the following two ways: (a) we use a different approach for collision detection, and (b) we use a different numerical optimization scheme. We note that there are variants of CHOMP that use gradient-free, stochastic optimization, including STOMP (Stochastic Trajectory Optimization for Motion Planning) (Kalakrishnan et al, 2011) and ITOMP (Incremental Trajectory Optimization) for real-time replanning in dynamic environments (Park et al, 2012).…”

“…All algorithms were run using default parameters and post-processed by the default smoother and shortcutting algorithm used by MoveIt!. We also compared TrajOpt to a recent implementation of CHOMP (Zucker et al, 2012) on the arm planning problems. We did not use CHOMP for the full-body planning problems because they were not supported in the available implementation.…”

Sampling-Based Methods for Motion Planning with Constraints

“…In recent years, approaches that use penalty functions and optimize over full trajectories have been proposed (20)(21)(22)(23). These approaches relax the hard constraints of the task (those that must be satisfied exactly, e.g., geometric task constraints and obstacle avoidance) into soft constraints (corresponding to some cost to optimize), combining the constraints into the formulation of the penalty function.…”