I think Covariant could be towards something similar. Lyft Level 5 makes an entire autonomous vehicle from other platforms. Those platforms have to realize their data is potentially useful on multiple platforms.
In my anecdotal small experiments I couldn't manage to get an agent trained on platformer 1 to do well on platformer 2. Transfer learning may be one way to deal with the ridiculous costs of RL but yeah seems like only Covariant is thinking of solving this problem
It's shocking how few players there are in such a massive space - outside of self driving cars have you been excited about any new entrants?
I think Covariant could be towards something similar. Lyft Level 5 makes an entire autonomous vehicle from other platforms. Those platforms have to realize their data is potentially useful on multiple platforms.
In my anecdotal small experiments I couldn't manage to get an agent trained on platformer 1 to do well on platformer 2. Transfer learning may be one way to deal with the ridiculous costs of RL but yeah seems like only Covariant is thinking of solving this problem