>>13686744Thanks, I will take 100 of the final version.
It's cheaper and better than any rover JPL has made so far and I won't have it jump over stuff.
>>13686749I did say it was dumb but I wasn't the smug guy.
>second one does not use any neural networks.It uses for vision and body control. Anon proposed a NN to simulate the robot moving around, not that different (but pointless and dumb).
>they told it what path to go and it figures it out using classical control. then brute force your way through with dozens of crashes and small adjustments until you get that acrobatic routine.If that's true just don't do acrobatics then.
>tell that robot to take a shovel and dig a hole without falling over 20 times. tell it to use a screwdriver to put together some valve or electrical component. literally harder than making a car drive itself.Easier than training it for acrobatics. For digging use a excavator instead.
Best solution is to send people and fixing lag with NNs is dumb.
