In 1996 I attended the IROS robotics conference in Osaka, Japan as part of a group of grad student robotics researchers. We toured the robotics labs at Tokyo University (Todai). There were a lot of student projects in humanoid robots.
Here's a video of a recent humanoid robot project. It's similar to the student projects I saw, but a bit more advanced. Aiko can read (apparently) and visually track objects, react to touch, and process some natural language. Note the command language the inventor uses to control the robot: "Aiko, Japanese mode." "Aiko, trace object." These are verbal equivalents to pushing buttons on a control panel.
The entire effect of the lifelike mask, loud mechanical motors whirring, and stilted conversation is a little creepy. I think this demo is relevant to the "Uncanny Valley" hypothesis. I know it's a popular idea, that if we make our machines more "lifelike" then people will trust them and accept them, but I'm not seeing any evidence for that. I mean, look how things turned out in Blade Runner.