Communication system for robots works better with longer tests in 3D simulation
Evaluating the Transfer of Co-Evolved Communication from 2D to 3D Simulation
RoboticsNeural and Evolutionary Computing
Summary
This paper looks at how a communication method developed between two robots in a simple 2D virtual world performs when used in a more realistic 3D simulation with physics. The authors found that giving the robots more time to complete tasks improves their success rates and that the communication mainly helps avoid getting stuck rather than guiding movement directly. They also studied how differences in robot roles affected communication and ruled out some technical glitches as causes. Despite this, the method doesn't work perfectly in 3D yet, and the authors discuss why that might be.
What this means in practice
- •For robotics developers: Improve cooperative robot communications by tuning time budgets and role asymmetries when moving from 2D prototypes to physical-world simulations.
- •For simulation engineers: Design evaluations with longer episodes in physics-based 3D simulators to better measure multi-agent system performance transferred from 2D setups.
Authors
Fernando Montes-Gonzalez
Abstract
This work examines the transfer of a co-evolved communication mechanism between two robotic agents from a discrete two-dimensional (2D) simulator to a three-dimensional simulator with real physics (3D). The study focuses on whether a communication mechanism co-evolved in a 2D environment retains its functional role after transfer to a 3D physics-based simulator. To support this analysis, the effects of the episode time budget, the social cue, and the asymmetry between the two co-evolved roles were examined. The results indicate that the success rate increased approximately linearly with the evaluated time budgets, with no evidence of a plateau between 2,000 and 6,000 physics steps, suggesting that evaluations based on shorter episodes may underestimate the performance of the trained controllers. In both simulators, the social cue functioned primarily as a jam- assistance mechanism rather than as a navigation guide, although with a more pronounced effect in 2D. Analysis of eight independent evolutionary runs revealed a consistent direction of asymmetry, although its magnitude varied across runs. Controlling the processing order between agents allowed us to rule out an artifact of the physics engine. Finally, the results are discussed in terms of the factors that may contribute to the remaining performance gap observed after transfer.