yeah proper robots is absurdly expensive. But its also multi-faceted. So if your interest is in machine vision, you can do alot of experiments with a web cam (which are really good now!)

I built an optical feedback robot arm for under $100 [1]. At work I use a robot where an individual sensor costs $100,000

The algorithms are much the same though. Doing control on cheap hardware is more challenging, so if you can do something on cheap hardware, you really have done something noteworthy! Expensive hardware pays for reduction of noise, it treats the symptoms of reality but does not solve them.

To truly solve the robotics issue of environmental noise, you have to solve it in software which is really hard but can be investigated cheap on cheap hardware. Robot projects fail because of software, not hardware normally. You should always build the software stack before spending lots of money on hardware.

[1] http://edinburghhacklab.com/2013/05/robotics-adaptive-contro...

Here are some cheap ideas:

1. Improve the robot arm control I outlined above. There is a huge market for cheap robot arms you could scale the ideas prototyped in [1] upto.

2. Active face recognition, put a camera with a zoom on a pan and tilt. Try to zoom into peoples faces for better recognition. Animals exploit shitty eye hardware with active attention in a way our current robots don't exploit.

3. voice activated inventory system for electronic components. Every hackerspace can't find the components they own. Popout a component tray via a push gantry behind it.

4. 3D hardware localization system. Actually determining the position of something is hard. Its the building block for gather training sets to do more challenging problems