The Problem
REVOLVE has worked reliably for the modular robot based on Robogen, but what about other robot morphologies that use different components? Evolutionary / co-design algorithms need to be tested on a variety of domains/tasks to fully evaluate their performance, currently, REVOLVE only supports one domain.
The Proposed Solution
More modular robots should be added.
OpenAI's gym already provides several mujoco-based robotic environments (ant, half cheetah, hopper, swimmer, etc) that are widely used within the RL community.
Documenting how to implement additional domains will also prove useful for others who try to utilize REVOLVE for their specific robot type.
The Problem
REVOLVE has worked reliably for the modular robot based on Robogen, but what about other robot morphologies that use different components? Evolutionary / co-design algorithms need to be tested on a variety of domains/tasks to fully evaluate their performance, currently, REVOLVE only supports one domain.
The Proposed Solution
More modular robots should be added.
OpenAI's gym already provides several mujoco-based robotic environments (ant, half cheetah, hopper, swimmer, etc) that are widely used within the RL community.
Documenting how to implement additional domains will also prove useful for others who try to utilize REVOLVE for their specific robot type.