Leader-follower formation enabled by pressure sensing in free-swimming undulatory robotic fish

Abstract: Fish use their lateral lines to sense flows and pressure gradients, enabling them to detect nearby objects and organisms. Towards replicating this capability, we demonstrated successful leader-follower formation swimming using flow pressure sensing in our undulatory robotic fish ($\mu$Bot/MUBot). The follower $\mu$Bot is equipped at its head with bilateral pressure sensors to detect signals excited by both its own and the leader's movements. First, using experiments with static formations between an undulating leader and a stationary follower, we determined the formation that resulted in strong pressure variations measured by the follower. This formation was then selected as the desired formation in free swimming for obtaining an expert policy. Next, a long short-term memory neural network was used as the control policy that maps the pressure signals along with the robot motor commands and the Euler angles (measured by the onboard IMU) to the steering command. The policy was trained to imitate the expert policy using behavior cloning and Dataset Aggregation (DAgger). The results show that with merely two bilateral pressure sensors and less than one hour of training data, the follower effectively tracked the leader within distances of up to 200 mm (= 1 body length) while swimming at speeds of 155 mm/s (= 0.8 body lengths/s). This work highlights the potential of fish-inspired robots to effectively navigate fluid environments and achieve formation swimming through the use of flow pressure feedback.

• The paper presents a method for enabling leader-follower formation in free-swimming robotic fish (µBots) using artificial lateral line-like pressure sensing and an LSTM neural network for control.
• Experimental results show a follower µBot equipped with only two pressure sensors can effectively maintain formation with a leader within a one body length range after training a neural model efficiently with limited data.
• This research demonstrates the viability of simple flow pressure feedback and neural networks for sophisticated fluid navigation tasks, opening possibilities for advancements in underwater swarming robotics and AI integration.

Leader-Follower Formation in Robotic Fish with Pressure Sensing

The paper presents a study on leveraging pressure-based flow sensing to enable leader-follower formation in free-swimming robotic fish, termed µBots. A key innovation lies in utilizing artificial lateral line-like pressure sensors to facilitate hydrodynamic perception in these bio-inspired robots. The authors outline a systematic approach to develop an expert policy for the robotic fish, combining experimental findings and machine learning techniques.

Methodology

The research utilizes undulatory fluid dynamics principles, aiming to replicate the hydrodynamic interactions observed in actual fish schools. The follower robotic fish is equipped with bilateral pressure sensors and an Inertial Measurement Unit (IMU) to detect changes in hydrodynamic pressures caused by the leader. The study starts by identifying efficient static formations through experiments, determining optimal positions for sensor detection of hydrodynamic variations.

An Long Short-Term Memory (LSTM) neural network model was used to train a behavioral cloning policy, mapping real-time sensor inputs to steering actions. The model training emphasized efficiency by deploying data aggregation and imitation learning strategies, reducing both time and data requirements for policy development. This enabled the implementation of a practical controller with limited computational resources and data—the training lasted less than an hour and successfully managed to track a leading fish within defined spatial constraints.

Results and Implications

The experimental outcomes demonstrate that the follower µBot, equipped with just two pressure sensors, could effectively maintain formation with a leader within a one body length range. This indicates the viability of utilizing flow pressure feedback and relatively simple neural network models to achieve sophisticated fluid navigation tasks in robotic configurations.

The study introduces possibilities for advancements in underwater robotics, particularly in swarming and schooling applications, where nuanced flow sensing can optimize energy efficiency and adaptability to dynamic environments. Additionally, the integration of neural networks for control tasks underscores the growing intersection of AI and robotics, demonstrating a tangible solution for real-time control systems with practical constraints.

Future Directions

Potential future improvements could involve integrating larger arrays of pressure sensors, enhancing the pattern resolution and robustness of flow feedback perception. Similarly, extending the scope of the neural model to accommodate adaptive behaviors, such as speed adjustments or multi-fish interactions, could further enhance the utility and sophistication of robotic fish in complex aquatic settings.

The findings and methodologies of this paper contribute significantly to the fields of bio-inspired robotics and underwater sensing technology, illustrating the potential for synergistic advancements using biologically inspired strategies and contemporary machine learning frameworks. This work not only advances our understanding of robotic engineering but also opens new avenues for leveraging fluid dynamics in broader autonomous navigational applications.