I can see this being a big benefit to those with minimal mobility. Not just those limited to voice command, but also to those with limited muscle control. I know a gentleman who had a stroke, and can't talk. His hand movements are also pretty limited, and a bit shaky. Such a robot can interpret his hand controls on a joystick, and interpret them while filtering out his unintended movements.
Some years ago spoke briefly with one lady who was researching ways for severely handicapped children to be able to play. She was saying that even a tiny bit of ability to control the world around them could be very valuable. For your project, even if the wheelchair is of limited use (such as inside the house), I imagine that having just a little more control over their lives will be greatly appreciated.
Of course the biggest issue with making a device to assist a handicapped person is that you are taking some responsibility for that person's safety. Of course, your efforts are an academic exercise, so you won't truly have that burden. But you should be working as though that responsibility were there.
I think your biggest challenge is not getting the chair to move to a command. Rather, it is being absolutely sure the chair cannot go down a step or flight of stairs. Bumping into things is less of a safety issue, but you certainly don't want the person to feel like a bull in a china shop.
If the chair is restricted to being inside a controlled environment (such as an individual's house or a nursing home) then you can take advantage of that situation. The person will probably have only a few standard places where they go, and common paths they will take between those places. E.g. the Living Room, The TV, By the Window, At the computer desk.
Then, the user only has to indicate their intended destination, and the chair's task it to follow a pre-programmed route. If the person wants to go somewhere off those routes, then the interface needs to be more complicated to allow the person time to make sure they are not knocking stuff over.
On the pre-programmed routes the user can verbally stop the chair themselves, though it should be kept clear most of the time. But I think the chair should still monitor that the path is clear, especially if it needs to back up at any time.
Knowing the difficulty robots have with obstacle avoidance, I would work to put as much responsibility for avoiding obstacles on the user as you can. Perhaps a video camera to view what is behind them so they can back up with confidence. When moving forward, a laser might be able to scan out the planned path, so the user can see if the path is what they really want to do, and that the path is clear.
Joe Dunfee