Synchrono: A simulation framework for machine learning in multi-robot applications.

We present here a simulation infrastructure designed to allow for safe, low-cost, and rapid development, testing, and evaluation of robot control strategies particularly in scenarios that are difficult or impossible to test in reality. The software platform provides (a) simulated dynamics to support interaction between the robots and the environment, (b) simulated sensing to provide the robots with a realistic perspective of the environment, (c) simulated communication to support inter-robot communication used for collaboration and coordination. Support of dynamics allows the exploration of physics-limited scenarios such as slip, collisions, non-linear flexibility, or interaction with fluid. Combining the dynamic simulation with virtual sensing allows us to enhance the capability of the control strategies by exploring sensing limited scenarios such as low light or inclement weather. In addition to dynamics and sensing, simulated communication in the virtual environment allows for coordination that may be used in multi-robot scenarios. These complex scenarios that include dynamics, sensing, and communication are parallelized in this simulation framework such that hundreds of variations can be performed to probe difficult or impossible to test edge-cases.