University of Twente Student Theses

Login

Design of a direct collocation model predictive control framework for legged locomotion based on whole-body dynamics and explicit contact phases

Roos, R.A. (2020) Design of a direct collocation model predictive control framework for legged locomotion based on whole-body dynamics and explicit contact phases.

[img] PDF
7MB
Abstract:To increase the walking performance of active exoskeletons, and legged robots in general, model predictive techniques can provide more dynamic and robust control. Model predictive control (MPC) can optimize a trajectory and corresponding inputs for a given model, in a given setting. MPC has been used for legged locomotion before, though previous works tend to use simpli�ed dynamics and therefore do not optimize the robot joint torques. In this work we set out to design an MPC framework for legged locomotion that includes whole-body dynamics, such that trajectories can be generated that are e�cient in joint torque. MPC based on implicit contact would be preferable, but showed infeasible with the direct collocation method. Instead an approach with explicit contact, based on prede�ned stance and swing phases, was used. This is an extension of TOWR [1], with the addition of whole-body dynamics. MuJoCo was used to model the system dynamics and IPOPT as numerical optimizer. The framework proved e�ective in locomotion generation and versatile to robot models, terrains and gaits. For a wide variety of robot types gait trajectories are optimized. However, the optimized trajectories are ine�ective when applied in an MPC simulation.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Programme:Electrical Engineering MSc (60353)
Link to this item:http://purl.utwente.nl/essays/85467
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page