Humanoid robots are taking too big of a step and are ridiculous. If they have good resilience, they might succeed, but it would be a thankless task. In my view, the final product form will definitely be part AI and part human remote control. One remote control operator will manage multiple robots to serve the boss. Currently, AI based on neural networks is not reliable or stable, and it should not be used to solve practical production problems.
If a tool robotic dog falls on the road, you would think it's garbage because you bought it hoping it would help you with grocery shopping; it has to have functionality. But if your pet dog trips while walking, you would find it very cute. Using AI to create pet robots for emotional support is the reasonable path. AI is smarter than cats and dogs.
Making the appearance of pet robots cuter, fluffier, and warmer than cats and dogs would make them appealing to touch. Their movements should be agile. Using large models to give them a unique soul, remembering everything from interactions with their owners, and even customizing their personalities while retaining personality data would be ideal. Unfortunately, no such company has been found yet. The closest companies are those making AI toys, primarily selling to children.
Pet robots are definitely the next global trend. When AI technology develops to have true thinking abilities and spatial understanding, it will become reliable for robots to perform tasks like screwing.