Neuromorphic Control for Real-Time Obstacle Avoidance in Underwater Robotics (In collaboration with the Soft Robotics Lab, ETH Zurich)

alternate text

The underwater environment presents a unique set of challenges for autonomous systems: limited visibility, low-bandwidth communication, unpredictable fluid dynamics, and severe energy constraints. Biological organisms such as fish and dolphins solve these challenges with remarkable efficiency by relying on energy-frugal sensing, fast reflexes, and adaptive motor control—all integrated into compact neural architectures. Inspired by these biological systems, neuromorphic computing offers a promising technological paradigm to r eplicate similar capabilities in robotic platforms.

Neuromorphic processors, such as the DYNAP-SE developed at the Institute of Neuroinformatics (UZH-ETH), mimic the structure and dynamics of biological neural circuits. While neuromorphic solutions have shown promise in vision-based robotics, their application to underwater robotic platforms remains largely unexplored.

This project builds upon prior work in which a robotic fish platform [1] was equipped with ultrasound sensors and interfaced with a neuromorphic processor to perceive its surroundings. The goal of this thesis is to design, implement, and validate a fully neuromorphic control stack for underwater obstacle avoidance. The student will model a spiking neural network (SNN) that processes sensory inputs from the ultrasound sensors and generates motor commands for the fish’s tail actuation. This sensorimotor loop will be implemented on the DYNAP-SE neuromorphic processor, enabling closed-loop, on-chip control without reliance on external computing.

 

Expected Outcomes

The expected outcome of this thesis is a functional demonstration of a bio-inspired underwater robot performing real-time obstacle avoidance using neuromorphic hardware for both sensing and control. The project will contribute to the emerging field of embodied neuromorphic intelligence and demonstrate how adaptive, low-latency behavior can be achieved using spiking neural networks in dynamic environments.

 

Requirements

  • - Proficiency in Python programming

  • - Solid understanding of neural network models or signal processing

  • - Familiarity with PCB design is a plus, but not mandatory

 

Contact

Dr. Elisa Donati — elisa@ini.uzh.ch

References

[1] Katzschmann, R. K., DelPreto, J., MacCurdy, R., & Rus, D. (2018). Exploration of underwater life with an acoustically controlled soft robotic fish. Science Robotics, 3(16), eaar3449.

back
back