Efficient collective swimming by harnessing vortices through deep reinforcement learning (1802.02674v1)

Abstract: Fish in schooling formations navigate complex flow-fields replete with mechanical energy in the vortex wakes of their companions. Their schooling behaviour has been associated with evolutionary advantages including collective energy savings. How fish harvest energy from their complex fluid environment and the underlying physical mechanisms governing energy-extraction during collective swimming, is still unknown. Here we show that fish can improve their sustained propulsive efficiency by actively following, and judiciously intercepting, vortices in the wake of other swimmers. This swimming strategy leads to collective energy-savings and is revealed through the first ever combination of deep reinforcement learning with high-fidelity flow simulations. We find that a `smart-swimmer' can adapt its position and body deformation to synchronise with the momentum of the oncoming vortices, improving its average swimming-efficiency at no cost to the leader. The results show that fish may harvest energy deposited in vortices produced by their peers, and support the conjecture that swimming in formation is energetically advantageous. Moreover, this study demonstrates that deep reinforcement learning can produce navigation algorithms for complex flow-fields, with promising implications for energy savings in autonomous robotic swarms.

• The paper demonstrates that deep reinforcement learning enables smart swimmers to harness vortex wakes, achieving a 32% boost in efficiency.
• It shows that fish-like formations can reduce energy cost by 36% by aligning with oncoming vortices through adaptive control.
• The study integrates high-fidelity flow simulations with RL algorithms, providing insights for energy-efficient designs in robotics.

Efficient Collective Swimming by Harnessing Vortices Through Deep Reinforcement Learning

The paper "Efficient collective swimming by harnessing vortices through deep reinforcement learning" investigates the mechanisms by which fish in schooling formations improve propulsive efficiency. The paper uses an innovative combination of deep reinforcement learning (RL) and high-fidelity flow simulations. The research addresses a significant question in biology and engineering: how do fish in schools extract energy from the fluid dynamics around them?

The paper establishes that using deep RL, a 'smart-swimmer' can optimize its swimming behavior by interacting intelligently with the vortex wakes generated by leading fish. This interaction can lead to substantial energy savings, providing a real-world incentive for fish to swim in formation. The findings confirm the hypothesis that fish may take advantage of unsteady flow fields to lessen their energetic expenditure, cementing the theory that formation swimming offers biomechanical benefits.

Two-dimensional simulations demonstrate that a follower in the wake of a leader can autonomously adjust its position to align itself with the momentum of oncoming vortices using deep RL. The RL-trained algorithm allows the follower to shed less energy against flow-induced forces and, as shown in numerical experiments, the 'smart-swimmer’ achieved a 32% increase in swimming efficiency and a 36% reduction in the cost of transport (CoT) compared to solitary swimmers.

For three-dimensional cases, deploying feedback control algorithms to maintain position against vortex rings revealed a corresponding increase in efficiency. This paper showcases how autonomous control strategies derived from RL might find applications beyond biology, such as energy-efficient motion strategies for robotic swarms in fluid environments.

The implications of these results have practical engineering potentials, particularly in designing autonomous underwater vehicles (AUVs) and robotic swimmers for industry applications. Theoretical advancements include insights into the dynamic interactions between flexible body movements and fluid dynamics that can inspire new optimizations in various animal locomotion models.

In conclusion, this paper provides quantifiable, controlled experimental data supporting the understanding of collective swimming mechanics, reinforcing the notion of significant hydrodynamic advantages in schooling. The success of deep reinforcement learning in this complex, dynamic setting suggests that similar methodologies could advance further computational fields, particularly those involving real-time decision-making and complex fluid dynamics.