Lunar robots - AI guidance where low latency control is impossible

[Al_]

TL;DR

There is a one second delay sending signals to the moon, so fine control of a robot is slow. Using AI to guide the robot instead.

For controlling robots on the Moon, for mining or construction.

We have robots that can copy human movements from camera vision. Factory robots can perform simple tasks after being shown what to do. But first-person controlled robots don't work very well with a one-second delay each way.

I propose a variation on the first-person control robot. Instead of always directly controlling the arms and grippers, the operator gives more general guidance.

E.g. Identifies a bolt, using the robots vision, and issues the command to undo it. Perhaps even suggests to approach with a certain tool from a certain angle.

E.g. Directly and slowly, with the one-second lag, show the robot how to do a task, then command it to repeat at its own speed without the lag.

E.g. Show the robot a route to travel using the robots own vision camera, then say go. The robot does it own hazard avoidance in case something suddenly moves into the path

 

[Filip Larsen]

I propose a variation on the first-person control robot. Instead of always directly controlling the arms and grippers, the operator gives more general guidance.

Not to rain on your parade, but engineers have been working on increasing levels of autonomy ever since the first rovers on Moon and especially Mars.

Latest experiment has been with generative AI for autonomous path selection for driving towards a target:
https://www.jpl.nasa.gov/news/nasas-perseverance-rover-completes-first-ai-planned-drive-on-mars/

 

[sophiecentaur]

TL;DR: There is a one second delay sending signals to the moon, so fine control of a robot is slow. Using AI to guide the robot instead.

But first-person controlled robots don't work very well with a one-second delay each way.

If an autonomous taxi can be operated in a busy city then a remote system on the Moon would need to be at least as competent. In fact, in many ways, a robot, working on its own could likely have a much easier task when 'interference' from other autonomous systems, working nearby. But even then, a remote management AI system would be the best solution. Directions from Earth would be more like 'build a rail system between these two bases' and the local manager would be telling the local bots 'dig ten holes here here and here . . . .'

As long as the timescale of a total task is much longer than the feedback loop delay then stability is not a problem (even with a very basic feedback loop).