Related Experiment Video
Updated: Oct 10, 2025

Experimental Investigation of the Flow Structure over a Delta Wing Via Flow Visualization Methods
Published on: April 23, 2018
Learning efficient navigation in vortical flow fields
Peter Gunnarson1, Ioannis Mandralis1, Guido Novati2
1Graduate Aerospace Laboratories, California Institute of Technology, 1200 E California Blvd, Pasadena, CA, 91125, USA.
Reinforcement learning enables robots to navigate efficiently in ocean currents. A velocity-sensing approach proved surprisingly effective, achieving high success rates for underwater robotic navigation.
Area of Science:
- Robotics
- Artificial Intelligence
- Fluid Dynamics
Background:
- Efficient point-to-point navigation is crucial for robotic applications like ocean surveying.
- Robots often operate with limited environmental knowledge or face time-varying currents, hindering traditional optimal control methods.
Purpose of the Study:
- To apply a Reinforcement Learning (RL) algorithm for discovering time-efficient navigation policies for a fixed-speed swimmer in unsteady 2D flow fields.
- To evaluate the impact of different environmental cues on navigation success and efficiency.
Main Methods:
- Utilized a deep neural network that takes environmental cues as input to determine swimmer actions.
- Implemented a Remember and Forget Experience Replay mechanism within the RL algorithm.
- Tested both velocity sensing and bio-mimetic vorticity sensing approaches.
Main Results:
- Swimmers successfully exploited background flow to reach target locations.
- Navigation success was dependent on the environmental cue used.
- Velocity sensing significantly outperformed vorticity sensing, achieving near 100% success.
- The velocity sensing approach approached the time-efficiency of optimal navigation trajectories.
Conclusions:
- Reinforcement Learning can discover effective navigation policies for robots in complex flow fields.
- Velocity sensing is a more effective environmental cue than vorticity sensing for this navigation task.
- RL-based navigation offers a promising solution for autonomous underwater vehicles in dynamic environments.
Related Concept Videos
Turbulent Flow: Problem Solving
Temperature is a key factor in CO2 solubility. In this case, the CO2 gas and the liquid are cooled to 20°C. Lower temperatures...
Irrotational Flow
Velocity and Acceleration in Steady and Unsteady Flow
The acceleration can be generalized to any point in the flow, and expressed as components along three perpendicular directions, representing changes in velocity over...
Uniform Depth Channel Flow: Problem Solving
Bernoulli's Equation for Flow Normal to a Streamline
The pressure difference depends on the fluid's velocity and radius of curvature. The pressure variation is minimal in flows with nearly straight streamlines.
Bernoulli's Equation for Flow Along a Streamline

