Publikation
Towards Adaptive Neuromorphic Sensing based on Active Efficient Coding
Quentin Halbach; Vibhu Kaushik; Aheli Saha; Nguyen H. P. Nguyen; Arne Roennau; Chiara Bartolozzi; Alessio Franci; Elisabetta Chicca; René Schuster; Claudia Lenk
International Workshop on Active Inference (IWAI), 2026.
Zusammenfassung
Autonomous systems need a stable, efficient, and fast perception of their environments to react appropriately to dangers, navigate, and interact with others. Adaptation of sensing and processing is one approach to achieve this. We develop a neuromorphic system for perception, which uses three modalities, vision, hearing, and tactile, and implements active efficient coding of the sensory input. Thereby, the sensor properties, including encoding of the signal and the sampling procedure, e.g., movement of the tactile sensor over a surface, are adapted to yield efficient encoding and optimal performance in perception tasks like object detection. Here, we present results regarding the design of adaptive neuromorphic sensors, the identification of sensor parameters, which are most important for performance and efficiency metrics, the development of a controller to tune the sensor properties, and the system setup to merge the signals and processing of the different modalities.