Non-linear Model Predictive Control Based Motion Planning Among People
Authors
Jensen, Martin Bieber ; Callisen, Simon Nanoq ; Reiser, Balazs
Term
4. term
Education
Publication year
2020
Submitted on
2020-06-16
Pages
118
Abstract
Dette speciale udvikler en ruteplanlægger for en ikke-holonome mobilrobot baseret på ikke-lineær modelprædiktiv styring (NMPC). Ikke-holonome betyder, at robotten, ligesom en bil, ikke kan bevæge sig sidelæns. Målet er at gøre robotplatformen PAL moving base i stand til at navigere sikkert i et dynamisk miljø med både statiske og bevægelige forhindringer. Der opstilles en ikke-lineær kinematisk model af robotten og en enkel, lineær tilstandsrum-model af bevægelige forhindringer. Begge modeller bruges i en NMPC, som forudsiger, hvordan robot og forhindringer bevæger sig på kort sigt, og beregner styresignaler, der fører robotten mod sit mål uden kollisioner. Udformningen af planlæggeren blev drevet af parallelle simulationer af forskellige planlægningsmetoder. Det resulterede i en lokal NMPC-planlægger med fokus på setpunkt-stabilisering og med multiple-shooting diskretisering, en numerisk metode der opdeler tidshorisonten i delintervaller for lettere at håndtere ikke-lineariteter og begrænsninger. Derudover udvikles to metoder til at repræsentere polygonformede forhindringer, som indgår som begrænsninger i NMPC’en, så planlæggeren holder robotten i sikre områder. Planlæggeren implementeres på PAL moving base og sammenlignes med en moderne planlægger. Testene viser, at en sammenlignelig ydeevne kan opnås med NMPC-tilgangen, samtidig med at den kan håndtere bevægelige forhindringer.
This thesis develops a path planner for a non-holonomic mobile robot using nonlinear model predictive control (NMPC). Non-holonomic means the robot, like a car, cannot move sideways. The goal is to enable the PAL moving base robotic platform to navigate safely in dynamic environments with both static and moving obstacles. A nonlinear kinematic model is built for the robot, and a simple linear state-space model represents moving obstacles. These models are integrated into an NMPC that predicts short-term motion of both the robot and obstacles and computes control actions that guide the robot to its goal while avoiding collisions. The design was guided by parallel simulations comparing different planning methods. The final system is a local NMPC planner focused on set-point stabilization, using multiple shooting discretization—a numerical technique that splits the prediction horizon into segments to better handle nonlinearities and constraints. In addition, two methods for representing polygon-shaped obstacles were developed and added as constraints, ensuring the optimizer keeps the robot within safe regions. The planner was implemented on the PAL moving base and compared with a contemporary planner. Tests indicate that comparable performance is achievable with the NMPC-based approach, with the added ability to account for moving obstacles.
[This abstract was generated with the help of AI]
Keywords
Documents