An Obstacle-Edging Reflex for an Autonomous Lawnmower

K.A. Daltorio, A.D. Rolin, J.A. Beno, B.E. Hughes, A. Schepelmann, M.S. Branicky, R.D. Quinn, J.M. Green

Abstract:	We developed a controller that allows our prototype lawnmower, CWRU Cutter, 1st place winner of the 2009 ION Autonomous Lawnmower Competition, to follow pre-defined paths and reflexively edge around obstacles before returning to the path. CWRU Cutter is equipped with localization sensors (GPS) and obstacle detecting sensors (LIDAR and camera). The best human lawnmower drivers mow parallel lines, interrupted smoothly as necessary to follow the contours of obstacles. To get this quality of cut from an autonomous mower, we developed a low-level obstacle avoidance reflex that keeps the robot close to obstacles when required to circumvent them. Our method requires less processing than common planning solutions for two reasons. (1) The environment is represented by a 1- dimensional Polar Freespace array rather than a 3-dimensional (x,y,theta) configuration space. (2) The robot only plans on the local environment recently perceived rather than planning all the way to the goal. Other methods for local obstacle avoidance often assume cylindrical and/or holonomic robots. These are not good assumptions for our rectangular, wheeled lawnmower. Instead, we pre-calculate the polar ranges associated with the robot’s footprint and the areas crossed by the footprint during constant curvature stops. These swept area ranges are easily compared with the Polar Freespace array that represents the environment. First, the robot uses GPS to generate initial velocity and angular velocity commands that steer the robot to a path of parallel lines covering the field to be mown. Data from a camera and LIDAR go into a 1-dimensional array of ranges (the Polar Freespace) that represents the local environment. If the initial command puts the robot on a collision course, the swept area ranges will be greater than the Polar Freespace ranges. To avoid obstacles, a reflex searches the velocity – angular-velocity space to find a safe command reachable from the current speed and as close as possible to the initial path command. Key reflex search parameters are examined in a MATLAB simulation assuming perfect localization and LIDAR data. A velocity-dependent extension factor is calculated that allows obstacle avoidance as opposed to halting in front of obstacles. The search resolution is adjusted to trade-off calculation speed and clearance. For example, by checking an average of 5 velocity/ angular-velocity command pairs per time-step (maximum of 14 checked commands per time-step) our 66cm by 100cm robot skirts a 2m-diameter obstacle with 1.3cm clearance. To compare, if we had used a previously published algorithm, such as Curvature-Velocity Method, that assumes the footprint was circular, the robot would not be able to pass any nearer than 54cm to the side of obstacles. We tested this controller on our robot, CWRU Cutter, with the blades on. The software was written in LABVIEW and running on an NI-cRIO at 10 Hz. We observed that the mower was able to edge boxes, soccer balls, and picket fences with side clearance of a few centimeters, and then smoothly return to following parallel paths after passing the obstacle. When rapid changes in the environment occur, such as a person walking in front of the mower, the robot stops until the environment stabilizes. In the future this method could be extended to allow the robot to backup and try again with relaxed clearances if the robot becomes stuck in a tight corner. Alternatively, these reflexes could be adapted for low-level safety checking on other non-holonomic non-cylindrical robots operating in cluttered environments.
Published in:	Proceedings of IEEE/ION PLANS 2010 May 4 - 6, 2010 Renaissance Esmeralda Resort & Spa Indian Wells, CA
Pages:	1079 - 1092
Cite this article:	Daltorio, K.A., Rolin, A.D., Beno, J.A., Hughes, B.E., Schepelmann, A., Branicky, M.S., Quinn, R.D., Green, J.M., "An Obstacle-Edging Reflex for an Autonomous Lawnmower," Proceedings of IEEE/ION PLANS 2010, Indian Wells, CA, May 2010, pp. 1079-1092. https://doi.org/10.1109/PLANS.2010.5507339
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In