According to Facebook, PyRobot will allow developers to get up and running with a robot quickly, thanks to it providing a higher-level abstraction on top of the robot operating system (ROS).
Similarly named as PyTorch, Facebook machine learning framework for Python, PyRobot will play a similar role to PyTorch in the arena of robot programming, Facebook scientists Abhinav Gupta and Saurabh Gupta say.
PyRobot will make it easier for the research community to use robotics data sets, algorithm implementations, and models, and also help them set benchmarks, compare their work, and build on each others’ results.
Facebook scientists claim PyRobot can bring down the time required to set up a robot from days to just a few hours. What makes things even harder in robot programming is that supporting multiple robots usually adds a significant complexity. Indeed, it is not easy to port code for one robot platform to another due to differences in API and available abstractions. Therefore, one of the goals of PyRobot is providing cross-robot abstractions, so you only need to import the right robot class from PyRobot while leaving the rest of your code unchanged to target a specific robot:
from pyrobot import Robot
robot = Robot('locobot')
target_joints = [
[0.400, 0.721, -0.471, -1.4, 0.920],
[-0.675, 0, 0.23, 1, -0.70]
]
robot.arm.go_home()
for join in target_joints :
robot.arm.set_joint_positions(joint, plan=False)
robot.arm.go_home()
One big advantage of this approach, say Facebook scientists, is that you can develop your robot application using low-cost robots such as LoCoBot, without locking yourself in. This would make it more affordable to deploy a fleet of robots that will operate and learn in parallel and allow you to improve your solution.
PyRobot provides abstractions to control joint position, velocity, and torque, for path planning, forward and inverse motion, and more. Currently, PyRobot supports two robot platforms, LoCoBot and Sawyer, with more coming. Foreseeably, PyRobot success and adoption will largely depend on how well Facebook and the PyRobot community will work to extend the number of supported robot platforms. Next on Facebook's roadmap is Universal Robots, but Facebook is already calling for the participation of the research community to provide support for more hardware.
PyRobot's approach to cross-robot development is different to that fostered by, e.g., H-ROS SoM, a modularity stack meant to be embedded inside modular robots so they provide the same programmable interface.
It is worth noting that both Amazon and Microsoft recently announced support for ROS in some of their products. Amazon introduced AWS RoboMaker, which is a cloud-service to develop, test, and deploy ROS-based applications, while Microsoft is working on integrating ROS into Windows 10. In fact, both solutions address a different set of concerns than PyRobot's and should be not confused with the latter.