Environments = Access to RL

We started Gridworld because we wanted to make RL useful for more domains. In research labs, the environment is usually some simple game clone where algorithms are tested on. The main discourse centres how well the algorithm learns, with the average score used as a proxy to gauge that performance. We think this can apply to real-world industry use cases, but with the main metric being the average score rather than algorithm effectiveness.

We are Oxford MEng graduates who did RL research at school. Yuhe was supervised by Prof. Jakob Foerster at FLAIR and researched opponent shaping in multi-agent RL. John was at the Quantum Devices Lab where he studied the application of RL to quantum chip design.

Both of us believe that RL will transform what performance and efficiency looks like in every sector. Access to this is bottlenecked by the lack of ready environments for a wide variety of domains. This is where we come in: (1) we build custom environments for your specific use cases and then (2) run RL on them to find the best possible operating solutions for you!

Reach us here.