Washing dishes

imagine ( a wide image of ) robot hands designed specially for washing dishes, with modular components, and servos control the joints. There are two arms, one has a brush, and other is hand like to hold utensils with good firm grip. It washes dishes automatically and places above the drying rack above. It also turns on the tap and off by itself (digitized mechanism for water inlet or can also handle tap on-off)

the only way robot can work is if its aware of its surroundings, and for it, it would need compact representation of information about its surroundings.

A 3d scene is seen as a set, with each temporally static (rigid) object as object itself with its orientation in 3d space, and if object deforms or gets modified, the parametrization of it following the object (e.g. genesis sim). And robot needs aware of it to dynamically act.

Training a random network can work and could be dumb as it would still require programming of rules, that are required for human to learn anyways, from data, and such training data must be included in training of the robot. Lets call this robot school, or robot learning (i.e. machine learning).

Now another challenge is its manufacturability, as I’m a single person without any resources or money, I have only the idea, and design links that a person can get it done and make it open source dish washer.
Things needed:

Couple of SO100 arm kits(i.e. 2 arms) https://wiki.seeedstudio.com/lerobot_so100m/

GPU (capable of running RT2-X robotics transformer /openvla / or ACT policy) and intelligent system capable of embodied robotics (for scalability)

A 3d design for the gripper which actually is the brush. And another gripper which firmly holds the dish, bowl, glass, etc. (we start slow)

arm design needs to be modified for aesthetics and minimal required DOF’s

Cameras (sensors) for robot to be aware of its surroundings. It can ofc, be used with variety of possibilities (e.g. NeRF, scene modelling , initial scene modelling (with rest in simulation), but mostly realtime on-line feedback control).