Surface Coverage Robotics

There is a tremendous opportunity for robotics in applications like farming, mowing, snow removal, subsurface characterization, and mine detection and elimination. The operations are slow, tedious, expensive, and in some cases, hazardous for humans to perform. Automating the equipment can increase productivity and improve safety.

In each of these applications, a large vehicle systematically traverses a given area of the ground in a regular pattern to deploy sensors or engage on-board implements. In the latter case, the vehicle often modifies the environment (e.g., cuts a crop or pushes snow to the side). In order to automate such equipment, path planning algorithms are needed to compute a coverage pattern that ensures the entire ground area is processed efficiently. In some cases, it may suffice to execute the pattern in an "open-loop" fashion using a position sensor (such as GPS) to track the preplanned trajectories. In other cases, imaging sensors such as CCD cameras must be carried on-board to detect obstacles, monitor the operation, or to achieve greater precision in tracking.

Our current focus in coverage robotics is on the automation of a harvesting machine, DEMETER (see figure below). DEMETER is a NewHolland 2550 Windrower equipped for retrofitted for computer control of the speed and steering functions. DEMETER is equipped with a CCD camera and on-board computer for visually detecting the line between cut and uncut crop and steering the machine to efficiently cut a row of alfalfa crop. To date, DEMETER has cut alfalfa on a farm in Western Pennsylvania at a maximum speed of 4.5 mph. By May of 1996, we plan to cut 100 acres of crop fully autonomously on a farm in El Centro, California.

The algorithms for the first version of the DEMETER perception system were developed by my graduate student, Mark Ollis. For his Ph.D. thesis, Mark is developing adaptive algorithms for detecting and tracking boundaries typically encountered in farming, mowing, and snow removal operations. Mark is exploiting the fact that machines performing these operations create the boundaries that they later must track. The algorithms used for tracking the boundaries can be adapted and tuned based on data collected from sensors mounted on the rear of the machine observing the newly created boundaries. The figure below shows the current crop segmentation algorithm in operation. The points indicate the detected crop line boundary for each row in the image.

Back to home page.