Improving Local Trajectory Optimization by Enhanced Initialization and Global Guidance

Maximilian Kramer,Torsten Bertram

doi:10.1109/access.2022.3159233

Maximilian Kramer, Torsten Bertram

Open Access

https://doi.org/10.1109/access.2022.3159233

Copy DOI

Abstract

Trajectory optimization is a promising method for planning trajectories of robotic manipulators. With the increasing success of collaborative robots in dynamic environments, the demand for online planning methods grows and offers new opportunities as well as challenges for trajectory optimization. Special requirements in terms of real-time capabilities are one of the greatest difficulties. Optimizing a short planning horizon instead of an entire trajectory is one approach to reduce computation time, which nonetheless separates the optimality of local and global solutions. This contribution introduces, on the one hand, Extended Initialization as a new approach that reduces the risk of local minima and aims at improving the quality of the global trajectory. On the other hand, the particularly critical cases in which local solutions lead to standstills are mitigated by globally guiding local solutions. The evaluation performs four experiments with comparisons to Stochastic Trajectory Optimization for Motion Planning (STOMP) or Probabilistic Roadmap Method (PRM*) and demonstrates the effectiveness of both approaches.

Highlights

Trajectory optimization enables an intuitive way to apply motion specifications to trajectory planning for robotic manipulators
E.g., enlarging the feasible set by increasing the planning horizon or changing the cost function to provide a local solution without standstill
Since the first experiment demonstrated the effectiveness of solving a local problem initialized by LINEAR INTERPOLATION (L-IN) or MIRRORED INITIALIZATION (M-IN) instead of warm starting, this experiment investigates the effects on the closed-loop trajectory (CLT) when the main instance is initialized by parallel solutions according to Fig. 2

Summary

INTRODUCTION

Trajectory optimization enables an intuitive way to apply motion specifications to trajectory planning for robotic manipulators. The available planning time is limited and, depending on the dynamics of the environment, might be insufficient to perform trajectory optimization in a loop For this reason, there are measures to simplify the planning problem. VOLUME X, 2022 at the price of reduced quality of the solutions This contribution analyzes the impact of planning short local trajectories that simplify the planning problem and proposes methods to mitigate suboptimality. Interior point [14] or sequential quadratic programming [15] approaches have become standard and efficiently exploit the sparsity of collocation or multiple shooting [5] Their main weakness is their dependency on a proper initialization, which is a serious factor especially for nonlinear problems such as trajectory planning. Pure path planning methods usually do not fulfill the requirements of the online trajectory planning problem without further ado

PROBLEM DEFINITION

PARALLEL PLANNING AND ENHANCED

INITIALIZATION BY LINEAR INTERPOLATION (L-IN)

EVALUATION

EXPERIMENT II (SCENARIOS A + B)

CONCLUSION AND OUTLOOK