Materials: [ RobotConvexCorridorCaseStudy.pdf]
This video introduces the problem statement that frames the “nonlinear
predictive control” developments of video [
In here, we discuss the “big picture” of what moving a robot from a start configuration to a goal one means in a cluttered environment. We need:
Stating the problem correctly as a mathematical optimization problem: start/end configurations (or regions), defining a cost index summarising in one number the application requirements.
High-level planning (preliminary waypoints, safe corridor)
Actual numerical optimization, smoothing the preliminary waypoints within the safe corridor
Analysing the solution (check that we have a feasible solution, and that it is sensible in practice, animations)
Implementing the solution either in open-loop (just replay the control actions moving from waypoint to waypoint) or closed-loop (two options: either recompute the MPC optimal trajectory or converge to the precomputed one via an error-based control law plus feedforward)
Dealing with the sim-to-real gap: maps have errors, localization is imprecise, unexpected obstacles (maybe moving ones) must be handled. This can be addressed the hard way, i.e., remapping (SLAM), replanning (spatio-temporal planning if obstacles are moving), or, well, as a simplistic solution, adding a potential field generating an artificial repulsive force from obstacles and hoping that we don’t get stuck in, say, an “U-shaped” passage.
The next videos [
Alternatively, we may consider posing the collision avoidance problem instead
of staying inside a safe corridor as staying outside of polyhedral obstacles (or
polyhedral approximations thereof). This idea is discussed in depth in a
continuation video [
I acknowledge the collaboration of my colleagues J.L. Pitarch and L. Armesto in drafting the code and providing background and fruitful discussions on the case study.
*Link to my [ whole collection] of videos in English. Link to larger [ Colección completa] in Spanish.
Colección completa en ESPA
Previous Control óptimo (LQR) de procesos lineales con desigualdades matriciales lineales (LMI): teoría y ejemplo Matlab
Next Nonlinear predictive control of unicycle robot (2): code analysis, part I, preliminaries
Full collection in ENGLISH [View all]:
Previous Duality of force/manipulability ellipsoids in a robot (inverse behaviour)
Next Nonlinear predictive control of unicycle robot (2): code analysis, part I, preliminaries