Directed Research Presentation
Rishabh Singh
Perception and Autonomous Robotics Group (PeAR)
Racing the Wild: High-Speed Autonomous Flight Through Forests
Tuesday, Dec 12th 2023
5:00 PM to 6:00 PM
UH405
Abstract:
High-speed autonomy of drones is critical in life-saving scenarios such as rescue operations during forest fires or natural disasters. To this end, we employ a dual-pronged strategy to elevate drone autonomy. We meticulously examined inherent inefficiencies in commonly employed quadcopter designs, driving the development of a streamlined model with substantially reduced drag. This innovative design targets not only superior performance at higher speeds but also ensures extended flight times while maintaining extreme agility, thereby fortifying the drone's capabilities for swift and effective response in critical scenarios.
Concurrently, we also address the persistent challenge of reliable outdoor forest navigation using only onboard sensors, a particularly demanding task for small drones with limited payload capacity. To this end, we introduce EdgeFlowNet, a high-speed, low-latency dense optical flow method running on an Edge TPU used for making decisions. We showcase the efficacy of our approach in obstacle avoidance and navigation through narrow gaps that simulate dense environments, providing valuable insights into the effectiveness of our approach in dynamic, challenging settings.
Advisor:
Prof. Nitin J. Sanket