Integrating path planning and robot control
There is often a conflict between planning the path a robot should take to achieve a desired task (high-level control) and the motion control needed for the robot to follow this path (low-level control). The problem is that if you decouple the path planning from the robot control, you might end up with paths that are impossible for the robot to follow because of physical constraints. Fully coupling the high-level and low-level control would solve this problem, although such intricate controllers are typically difficult to design.
To solve these shortcomings, Conner et al. propose a hybrid control strategy that combines low-level and high-level control in a smart way. As a test case, they consider a scenario where a robot needs to reach a goal while avoiding obstacles. The robot has a non-trivial body shape and is nonholonomic, meaning that it can not turn on the spot. The approach they developed is shown in the figure below. Local control policies, showed by fennel-shapped sets with vector field arrows, are responsible for making the robot drive towards a local goal. These policies respect the low-level dynamics and kinematics of the robot. A set of control policies can then be followed sequentially to reach a desired high-level behavior. To find the best path, an abstract tree representing the transitions between control policies is used.
Experiments were done with a LAGR robot in a fully known environment and with visual localization using landmarks. Results show that the method is successful in safely guiding the nonholonomic robot to its goal in an obstacle prone environment and that disturbances do not require the robot to replan its course.