Neuromorphic controller for an autonomous robotic vehicle

The goal of the project is to establish a long-term collaboration between Laboratory of Robotic Systems in Minsk and Institute of Neuroinformatics in Zurich. The two labs share interest in neuronally inspired control for autonomous robots, capable to work in real-world environments. In particular, in this first joint project, we will develop a reactive neuromorphic controller for mobile robots and autonomous vehicles based on branching algorithms, proposed by the Minsk's group. The control will be based not on primitive pairs of causes and reactions, but on a set of intersecting functional chains that include distant goals, preconditions, effects, and hidden causes. This scheme allows us to implement various branching algorithms using neuronal modules. Using this paradigm, we will implement a search controller in neuromorphic hardware, provided by the Zurich's group. The neural branching architecture will control the mobile robot to search for a way out of a labyrinth. The developed architecture will be validated on a self-balancing robot chassis. Neuromorphic sensory and actuator modules will be implemented for the chassis. Thus, the controller will solve two tasks -- balancing the robot based on two-axis analog accelerometer AD203 and finding a way out of the labyrinth using ultrasonic and infrared rangefinder signals. The controller will feature a high speed of sensory data processing thanks to event-based nature and local computation of neuromorphic sensory and processing systems. This will increase the robot's robustness to external perturbations and dead zones of the sensors. Potential application of such controller is the autonomous navigation in constrained and cluttered industrial environments.

Participants: