Student: Mitul Magu
Committee: Dr. Yezhou Yang
Abstract:
This work presents a centralized approach for collision avoidance among a set of mobile robots that are moving towards their respective goals in a shared space. An encoder-decoder LSTM (Long Short-Term Memory) based trajectory prediction model is implemented to recognize and prevent a future collision from occurring. An overhead camera is setup for concurrently tracking multiple robots and dynamic obstacles. In this centralized system, although each robot navigates to its own goal without outside interference, a pause command is given when the motion model detects the future location of two robots to be less than a threshold. This method is most effective in a large factory or office environment where robots can easily be tracked using an overhead camera. Experiments are conducted in the Webots simulator and the collision avoidance rate at different speeds with the LSTM model is compared with a similar kalman filtering based strategy and a sensor based model.
Zoom Room: https://asu.zoom.us/j/83541568934
Presentation Time: 12:00-1:00 PM (Arizona Time)
