MOTION PLANNING WITH OBSTACLES AND DYNAMIC CONSTRAINTS (ROBOTICS)

Robots must be capable of executing a variety of tasks precisely and under changing conditions. For this reason, it is not sufficient to have simple robot controllers which can be programmed to perform specified tasks. The concept of motion planning is a first step towards solving this problem. The planner is analogous to the brain of the human being in many respects. It accepts as inputs the nature of the task to be executed and the description of the environement and uses it knowledge of the capabilities of the manipulator to generate a sequence of commands to be executed by the controller.; The primary objective of this study was to design a planner which could be successfully integrated with the ordinary robot controllers. The aim was to produce not only the spatial path which the manipulator must track to avoid obstacles, but the actual control signals that could be fed to the motors driving the manipulator. Additionally, the task must be performed in the most optimal manner. The problem could not be treated as a purely geometric path-planning problem since manipulator dynamics and other dynamic constraints had to be accounted for.; The overall problem was posed as an optimal control problem with control and state variable inequality constraints. The geometric constraints imposed by the necessity to avoid obstacles were handled by modeling the obstacles as a composition of analytical primitives. The continuous-time problem was converted to a discrete-time problem, and solved using a non-linear mathematical programming technique. Computational issues such as discretization effects and algorithmic efficiency were investigated. The planner was implemented on the Unimate PUMA manipulator in simulations.; The efficiency of the planner is directly related to its information on the response of the manipulator to inputs. A procedure was developed to accurately obtain the dynamic behaviour of the manipulator by representing the dynamics in closed-form Lagrange equations, and determining the dynamic coefficients from the measured input-output behaviour. This was implemented successfully in experiments on a computer controlled robot.; To handle unexpected disturbances or modeling errors, an adaptive feedback controller, based on linearized manipulator dynamics along the planned trajectory, was designed. A recursive parameter identification scheme was used to identify the model and update the feedback gains. The feedback controller was employed in conjunction with the commands from the planner, and exhibited satisfactory performance.