ACTIVE SENSING IN SENSOR–BASED MOTION PLANNING WITH DYNAMICS J.C. Alvarez A. Shkel V. Lumelsky Sys.Eng.& Automation Area Mech.& Aeros. Eng.Dept. Robotics Laboratory ECE Dept., Univ. of Oviedo Univ.of California-Irvine Univ. Wisconsin-Madison 33204 Gijón, Asturias, Spain Irvine, CA 92697, USA Madison, WI 53706, USA Abstract: This paper studies active sensing strategies for an autonomous wheeled mobile robot moving on a flat surface among unknown static obstacles. The robot is defined by its geometry, its dynamic motion equations and a control policy. The latter comes from a path planning algorithm which produces changing intermediate goal co- ordinates to pursue, and motoring signals based on local information. We want to know which is the (minimum) area we need to explore with the robot sensors in order to guar- antee that a certain intended motoring signal is safe. The proposed solution depends on a) the sensor system design, b) the robot actual velocity and c) the robot dynamics. All these factors must be taken into account for safety and efficiency reasons. The result is an adaptive scanning procedure based on the robot motion circumstances at every time. Keywords: sensing algorithm, sensor-based motion planning 1 INTRODUCTION Real-time sensor-based robot navigation deals with the motion planning problem when only a subset of the workspace is known at each instant [3]. Algorithms of motion planning with incomplete information produce changing intermediate goal coordinates to pursue, and the associate control problem is to generate motoring signals based on local information in order to make the robot move towards the goal. But three restrictions hinder implementation in practice: 1) robot motion capabilities, 2) sensors characteristics and 3) computation time limits (because control commands must be issued at a fast pace). These are unavoidable experimental limiting factors that are frequently ignored in the algorithms. It is within this framework that we address the following problem: which is the (minimum) area we need to explore with the robot sensors in order to guarantee the safety of a certain intended motoring signals?. The question is meaningful as it deals with sensorial throughput, which is the major bottleneck when computing motion commands on real-time. This paper describes a method to decide where to gather this local information about the robot sorroundings, in order to guarantee that a given control command is safe. It provides an algorithm which deals with sensorial throughput making a selective scanning based in the robot motion circumstances at every time. 2 SENSING, MOTION AND DYNAMICS For the moment, consider the problem of sensor-based navigation as one consisting of two sepa- rate problems: a geometric task of path generation (call it Path Planner), for the robot to move in the workspace filled with obstacles, and a control task (call it Controller), which generates motoring com- mands. The input information to the Path Planner is robot current coordinates, , and the description of the surrounding obstacles. Its output is an intermediate target point, , and a straight-line path segment that leads to it. The Controller input is the current state , current velocity vector , point , and the path segment from the Path Planner which the Controller is expected to execute, Figure 1. Safety and motion efficiency requires to take into account the system dynamics. Otherwise, conflicts may arise, Figure 2. As the robot arrives at point along the path , it decides on a new intermediate target, , and a straight-line path segment to it. Since it arrives at with a non-zero velocity, because of the system dynamics, it cannot make a sharp turn suggested by the Path Planner. To preserve continuity