[smiley1437]

I remember ago Lars Blackmore of SpaceX released a paper on soft landing Falcon 9, that's the first time I'd encountered convex optimization

https://www.semanticscholar.org/paper/Lossless-Convexificati...

It blew my mind that you could convexify non-convex curves into useful-for-optimization convex curves to optimize for so many things simultaneously (physics constraints, control thruster limitations, sensor constraints, g forces, etc) and it's cool that part of the spectacular landings we get from SpaceX relies on it

[antman]

Non paywalled by Lars Blackmore http://www.larsblackmore.com/iee_tcst13.pdf

[PartiallyTyped]

til. For whatever reason I totally imagined it was some RL based method trained on sims. In my defense, RL is used for control problems as well, but this is so cool! Thank you for sharing.

[4gotunameagain]

no serious, safety critical system uses RL (except tesla "autopilot" and we see how that went). Control theory algorithms can be validated to work within the desired envelope and produce a valid solution.

The big advantage of convexifying the problem, is that when it is convex you have a guarantee it can be solved in fixed time, a major requirement for real time systems

[PartiallyTyped]

I wasn't thinking of DeepRL, but more on the more classical side of things with approximators other than neural NNs; but what you describe makes sense.

[bo1024]

On that side, reinforcement learning bleeds over into control theory, so you're partly right.