Title: Sensor-Based Autonomous Navigation for Mars Rovers: the "WedgeBug Algorithm" Authors: Sharon Laubach, Joel Burdick Abstract: We have developed algorithms for autonomous navigation of future Mars planetary rovers. These algorithms must be accurate, robust, and rely only upon onboard sensors. Furthermore, the algorithm leads to rover to its goal along the locally shortest path, while also minimizing the amount of sensing that is necessary to reach the goal. In this way, rover power is conserved.

Motivation & Aims

Future Mars exploration missions plan to make use of mobile robotic vehicles that are capable of traveling much farther distances (on the order of 10s of kilometers) than the Sojourner vehicle traversed during the last Pathfinder mission to Mars. Greater autonomy is clearly needed so that such vehicles can accomplish more between opportunities to communicate with Earth.

Research

Our effort has resulted in the development of rigorous motion planning algorithms which can navigate the rover from its current position to a given goal configuration that is specified by an earthbound operator in a variety of formats. We assume no a priori knowledge of the geometry of the nearby environment. The robot must rely only upon its onboard sensors (which consist of a stereo camera pair and a sun angle sensor) an onboard computing power to acquire all information and to compute and execute a successful motion plan. Because of severe power and computational constraints, we must also attempt to minimize the amount of the robot's unnecessary sensing and movement.

Our "WedgeBug" algorithm realizes all of these goals. It is a provably correct sensor-based motion planning algorithm that locally minimizes the rover's path length, while also minimizing the number of camera views that are taken and processed.

An important aspect of our work involves implementing our algorithms on the Rocky 7 prototype (see in the picture above) at the Jet Propulsion Laboratory, and testing/demonstrating their effectiveness in simulated Martian terrain.

Achievements

We have recently demonstrated complete multi-step autonomous navigation in simulated Martian terrain, as depicted in the figure below.

A sequence of false color depth maps produced from images taken by the Rocky7 Mars Rover prototype (seen in the above figure) during a 20 meter traverse in simulated Martian terrain at the Jet Propulsion Laboratory's "Mars Yard." Each "wedge" represents the field of view seen by the rover's stereo camera system. The color in each wedge represents terrain height. The green line represents the local path segments planned by the Rover. The sequence starts at the lower right. In each field of view, the rover plans a local path to edge of its field of view. After executing this subpath, the rover localises itself (the result of this procedure produces the "jags" in the path between each wedge). Then the sequence repeats until the goal is reach. The total traverse required approximately 57 minutes. By comparison, the Sojourner vehicle traversed a total of 100 meters in 1 month.

Selected Publications