Robot motion planning with visual constraints

Most current robotic systems treat robot motion planning and robot control as two independent problems handled by different subsystems: a motion planner that determines a collision-free path to achieve a task, and a robot control system that executes motion to follow the specified path. The decoupling of sensor-based robot control from motion planning may yield undesirable performance, because a feasible collision-free path may be very difficult to execute due to limitations in the sensors that support the control system.; A method that extends the configuration space framework to include sensor space constraints, especially visual constraints, is developed in this thesis. Some of the sensor space constraints, such as motion detectability/visibility, sensor singularity, and sensor optimum range can be considered during motion planning stage. Hence, the resulting path is not just feasible (collision-free) but also easy to control (deviation from the intended path can be easily recognized and corrected). A motion planning framework is developed to achieve this based on the concept of the Perceptual Control Manifold {dollar}({lcub}cal PCM{rcub}):{dollar} a manifold defined on the product of the robot configuration space and sensor space. Several different applications of the framework are described in this thesis.; Different sensors may have different behavior and limits to provide feedback. Especially for visual sensors, because of their flexibility, there are many possible sensor readings (image features) that can be chosen to provide feedback. A global measure that can be used to evaluate the performance of different image features for a visual servoing system is developed. Conversely, this measure can also be used to evaluate the goodness of different paths for given set of image features in a visual servoing system.; A computational approach to do motion planning on the {dollar}{lcub}cal PCM{rcub}{dollar} is developed. A modular multi-resolution grid representation is used to represent the {dollar}{lcub}cal PCM{rcub}.{dollar} Combined with a flexible multi-strategy search, this representation results in a hierarchical planner which is capable of generating plans for high degrees of freedom robot while considering sensor constraints.