Acous&c & Propriocep&ve Sensing on Outdoor Mobile Robots for Object Recogni&on Debadeepta Dey Hatem Alismail Jacqueline Libby
Main Idea Dirt Road Hard Road Gravel Short grass Acous&c Tall grass Small rock Mobile robot Large rock interacts with Classifiers Bush environment Pile of leaves Propriocep&ve ScaIered leaves Bramble Puddles Mud
PlaMorm
Example objects
Example approaches Feature Classifica&on Extrac&on Dirt Road Hard Road Raw Data Gravel • FFT Bag of words Short grass • Spectrogram Acous&c Tall grass • Signal Decay Small rock Noise Large rock filtering SVM Bush Proprio‐ • Sudden stops Pile of leaves cep&ve • Vibra&ons ScaIered leaves • Pitch change Bramble Neural Puddles Networks Mud
Related Work [1] Aliza Amsellem and Octavian Soldea. Func&on‐Based Classifica&on • from 3D Data and Audio. Pages 336–341. Proceedings of the IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, 2006. – Signal processing to classify func&onal components of objects • Peaks in FFT • Spectrogram • Signal Decay [2] Lauro Ojeda, Johann Borenstein, Gary Witus, and Robert Karlsen. • Terrain Characteriza&on and Classifica&on with a Mobile Robot. Journal of Field Robo&cs, 23(2):103–122, 2006. – Gravel, grass, sand, pavement, dirt – Gyros, accelerometers, encoders, motor current and voltage sensors, ultrasonic, infrared microphones – One NN for each modality, then compared – Gyros ‐> road type: gravel, pavement, or dirt – Microphones ‐> grass, not grass
Recommend
More recommend
