As the validators tests are the same as IDE’s slightly shifted, what will work in the IDE will work in the validator, so you can develop the physic system locally, use a genetic algorithm to find a local optimal path (like for the travelling salesman problem for exemple), and once you have this solution, hardcode it with tons of print(“angle thrust”)

I actually started by using my real time tweaking coefficients as genes and simulate the outcome.
After it’s a game of refining / redefining those coefficients and tweaking the algorithm itself.

What i can say is that pretty small changes can bring a lot of difference in the results.

Congratulations! I think I’ll have to rework my Mars Lander algo, I really pushed it to the limits already, not sure it can still get 5 more fuel points!