Perching failure detection and recovery with onboard sensing

摘要: Perching on a vertical surface carries the risk of severe damage to vehicle if maneuver fails, especially failure goes undetected. We present detection method using an onboard 3-axis accelerometer discriminate between perching success and failure. An analytical model was developed calculate acceleration differences for set decision times. Two distinct times were shown be effective, corresponding properly engaging gripper overloading gripper's capabilities. According machine learning feature selection algorithm, maximum Z axis quadrotor presence near-zero readings are most relevant features within these two time frames. Using features, algorithm discriminated with 91% accuracy at 40 ms, 94% 80 ms. Real-time recovery experiments 20 g verify method. improved approach that combines various correctly identified success/failure all trials average total falling distance 0.8m during recovery. discuss feasibility extending our other platforms.