![]() We began by formulating our rover as an agent in a Markov Decision Process (MDP) using OpenAI Gym so that we could model behaviors more easily for the context of reinforcement learning. Given a solid foundation in simulation, we turned to the algorithmic side of our project for testing. Communication between the simulation container and the host machine was handled by a custom socket interface. Docker allows for OS-agnostic dependency resolution and quick cluster deployment on cloud services such as Kubernetes. We also want to run these simulations on remote, GPU-enhanced servers for faster training and testing in the future. We needed a simulation that can be setup easily on different machines regardless of host environment. ![]() The one machine setup was fine for initial development but unscalable. The simulation is abstracted as an Open AI Gym environment so that algorithms interacting with it can be environment agnostic. ![]() The inputs to this Gazebo-ROS system are powers to the left and right wheels and the outputs are images, position, velocity, acceleration, and joint manipulation data. We model the rover using a URDF generated from a CAD model. We use ROS as a framework for connecting the simulation with logging and input. In Gazebo, we have simulation of the Moon running on one Linux machine. Over the last few months, the Autonomous Rover team’s software division has put together an end-to-end system for testing reinforcement learning algorithms in a custom environment, built with OpenAI Gym, ROS, and the Gazebo physics simulation engine. For more info, check out our project page! The end goal is to create a swarm of autonomous rovers that locate and extract resources on the Moon while exploring areas of the Moon that have not been studied in the past. The software team works on simulation of the moon environment and on training reinforcement learning algorithms. We employ reinforcement learning as a tool to minimize the cost of extracting minerals. For instance, rovers need to avoid mining in shadowed regions for extended periods of time, reroute for battery charging and storing collected resources, and communicate with each other to maximize coverage. However, extraction of these resources is incredibly difficult. ![]() Subsurface metals and water on the moon can be used to build tools for a habitable extraterrestrial environment, and can be used to refuel rockets. We are part of the larger AI Rover team, which under the guidance of NASA AMES is developing the algorithms that will be deployed on the Mechanical Rover team’s physical lunar rover intended for exploration and resource extraction.Īn autonomous rover system deployed on the moon could be used to mine resources. Our mission is to develop robust, artificially intelligent systems which can control a rover without human input. The Autonomous Rover team at STAC works on solving that problem. This vastly slows down exploration, and the dependence on human control limits the scale with which space agencies can deploy rovers. Every movement is decided upon by a human on Earth and transmitted through space – a 7-minute-long journey to reach a rover on Mars – before the rover can perform the motion. A resource rover is an extraterrestrial vehicle used for space exploration as well as resource extraction on various celestial bodies.Įxploration rovers are a staple of modern space exploration, but come with a major caveat: they are dependent on constant human control. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |