Sensing flow gradients is necessary for learning autonomous underwater navigation
- Yusheng Jiao 1, Haotian Hang 1, Josh Merel 2, Eva Kanso 3,4
- Yusheng Jiao 1, Haotian Hang 1, Josh Merel 2
- 1Department of Aerospace and Mechanical Engineering, University of Southern California, Los Angeles, CA, USA.
- 2Fauna Robotics, New York City, NY, USA.
- 3Department of Aerospace and Mechanical Engineering, University of Southern California, Los Angeles, CA, USA. Kanso@usc.edu.
- 4Department of Physics and Astronomy, University of Southern California, Los Angeles, CA, USA. Kanso@usc.edu.
- 0Department of Aerospace and Mechanical Engineering, University of Southern California, Los Angeles, CA, USA.
Related Experiment Videos
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.Robotic underwater navigation is improved by using egocentric observations and learning from trial-and-error. Sensing local flow gradients, not just velocities, is key for artificial swimmers to navigate without external references.
Area Of Science
- Robotics
- Biomimicry
- Fluid Dynamics
Background
- Robotic vehicles struggle with underwater navigation due to limited global positioning signals and complex flow dynamics.
- Aquatic animals exhibit superior underwater navigation capabilities, suggesting bio-inspired approaches are beneficial.
- Reinforcement learning offers a promising avenue for developing adaptive underwater navigation strategies.
Purpose Of The Study
- To investigate the feasibility of egocentric underwater navigation for artificial swimmers using only on-board sensors.
- To determine the necessary sensory information (flow velocities vs. gradients) for successful egocentric navigation.
- To explore the robustness and transferability of learned navigation policies in diverse flow environments.
Main Methods
- An artificial swimmer was trained using reinforcement learning to navigate to a destination in unsteady flows.
- The swimmer relied solely on egocentric observations from on-board flow sensors, without geocentric reference frames.
- The study compared navigation performance using only local flow velocities versus incorporating local flow gradients.
Main Results
- Sensing local flow velocities alone is insufficient for effective egocentric navigation.
- Incorporating local flow gradients is crucial for successful egocentric navigation in unsteady flows.
- Egocentric navigation strategies demonstrated rotational symmetry and enhanced robustness in unfamiliar flow conditions.
Conclusions
- Egocentric navigation in complex aquatic environments is feasible with appropriate sensory input (flow gradients).
- The findings support the hypothesis that aquatic organisms utilize flow sensors to detect gradients for navigation.
- This research provides a foundation for developing more capable, bio-inspired underwater robots and facilitates transfer learning for robot navigation.
Related Experiment Videos
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.