PhD Thesis

Kinodynamic Planning and Control of Closed-chain Robotic Systems

Work default illustration

Information

  • Started: 01/07/2016
  • Finished: 29/10/2021

Description

This work proposes a methodology for kinodynamic planning and trajectory control in robots with closed kinematic chains. The ability to plan trajectories is key in a robotic system. It provides a means to convert high-level task commands - like "move to location A", or "throw the object at such a speed" - into low-level controls to be followed by the actuators. In contrast to purely kinematic planners, which only generate collision-free paths in configuration space, kinodynamic planners compute state-space trajectories that also account for the dynamics and force limits of the robot. In doing so, the resulting motions are more realistic and exploit gravity, inertia, and centripetal forces to the benefit of the task. Existing kinodynamic planners are fairly general and can deal with complex problems, but they require the state coordinates to be independent. Therefore, they are hard to apply to robots with loop-closure constraints whose state space is not globally parameterizable. These constraints define a nonlinear manifold on which the trajectories must be confined, and they appear in many systems, like parallel robots, cooperative arms manipulating an object, or systems that keep multiple contacts with the environment. In this work, we propose three steps to generate optimal trajectories for such systems. In a first step, we determine a trajectory that avoids the collisions with obstacles and satisfies all kinodynamic constraints of the robot, including loop-closure constraints, the equations of motion, or any limits on the velocities or on the motor and constraint forces. This is achieved with a sampling-based planner that constructs local charts of the state space numerically, and with an efficient steering method based on linear quadratic regulators. In a second step, the trajectory is optimized according to a cost function of interest. To this end we introduce two new collocation methods for trajectory optimization. While current methods easily violate the kinematic constraints, those we propose satisfy these constraints along the obtained trajectories. During the execution of a task, however, the trajectory may be affected by unforeseen disturbances or model errors. That is why, in a third step, we propose two trajectory control methods for closed-chain robots. The first method enjoys global stability, but it can only control trajectories that avoid forward singularities. The second method, in contrast, has local stability, but allows these singularities to be traversed robustly. The combination of these three steps expands the range of systems in which motion planning can be successfully applied.

The work is under the scope of the following projects:

  • RobCab: Control strategies for cable-driven robot for low-gravity simulation (web)
  • KINODYN: Kinodynamic planning of efficient and agile robot motions (web)