Managing Limited Sensing Resources for Mobile Robots Obstacle Avoidance

Obstacle avoidance is a sensor-based task designed to steer a mobile robot towards intermediate goals while avoiding unexpected local obstacles in the robot’s surroundings. Responsiveness requires that the task is computed in short time intervals, fast enough to deal with the robot inertia and dynamics limitations. Simultaneously it has to guarantee detection by gathering enough information to allow the robot to react adequately to any potential danger. However, throughput and detection guarantee are opposite demands. If we try to assure the latter by augmenting the scanned area in the robot’s surroundings, it will increase the time taken to process the sensor data, proportionally decreasing the reactivity to the external world. On the other hand, scanning smaller areas can endanger the robot if some possible trajectories of motion are not swept by the sensors (e.g. the tunnel–vision problem). The previous trade-off can be addressed in different ways. Some obstacle avoidance algorithms are designed supposing that fixed sensorial information of the robot’s surroundings will be available when needed, e.g. range measures in a circular or rectangular robot-centered area (Lumelsky and Skewis, 1990), (Ulrich and Borenstein, 2001). Furthermore, others restrict the required information to the limits imposed by the robot’s actual motion, its dynamics and kinematics (Brock and Khatib, 2000). But the assumption that certain sensorial information will be available when required is not always realistic. For example, the particular field of view of conventional range sensors such as stereovision, sonar or laser, cannot adjust adequately to the specific perception requirements (Laubach and Burdick, 2000). Or the time needed for sensor data processing can have an impact on the real-time motion performance, making some sense modalities more recommendable than others (Kelly and Stentz, 1998). Finally, the availability of each specific sensing device can be an issue whenever the robot’s task solicitations exceed the robot’s computing capabilities, for example in robots with complex missions or with massive sensorial information. For these cases, treating sensing as an isolated phenomena leads to bottlenecks both oinn computational demand and in realtime performance. Our approach to the problem of making the required information available when needed is based on two ideas: 1) to define what the (minimum) information required is in order to guarantee the safety of motion for a given motion strategy, and 2) to use sensor management to obtain only the required info. Scanning only the essential area around the robot helps us to adapt to the real time demands about motion reactivity, and managing