Image or laser based obstacle detection/obstacle avoidance (ODOA) techniques often fail to steer a ground vehicle away from obstacles that are either not properly detected or their location estimates are inaccurate. In such cases it is necessary to detect collision, during or after it occurs, so as to allow for maneuvering of the vehicle around the obstacle. We are implementing two different algorithms for observing collisions with obstacles. The first approach is based on detection of high values of deceleration in the direction of motion. Although this technique is sufficient for cases that the vehicle engages a solid obstacle in high speed, it is not able to detect collisions, during slow motion, with elastic obstacles such as soft ground or vegetation. A second estimator will be developed for these cases. We intend to design and implement fault detection and identification algorithms that rely on vision based motion estimates and kinetic sensor measurements (IMU and/or wheel encoders) to infer immobility of the vehicle.
- Robot Image