spec me a programming project (robotics based)

Reaper 392

Reaper 392

Reaper 392

Soldato
Joined
15 Feb 2011
Posts
10,234
Location
Slough
for my third year group project in AI and cybernetics we have to program a robot torso using the Robot Operating System (ROS). as long as the project convenor is happy with the task we choose our project can be to get the robot torso to do pretty much anything we want it to do so long as we have the ability to program it. the end result will have to be fully automatic, i.e. we wont be able to control the arms ourselves, the robot will have to work out what to do for itself

since this leaves our choices wide open i was wondering if you guys have any fun ideas of stuff we could get the robot to do?


*extra info*

the robot currently has:
Two USB webcams to act as eyes, possibly to be replaced with firewire webcams

an Xbox kinect

one arm for precise control with a gripping hand (claw) at the end*
- this can not be back driven (ie, i cant move it myself)
- the hand is a claw that looks like the gripping devices used in chemistry lessons to hold test tubes

one arm for fast movement
- this can be back driven (ie, i can move it myself)
- currently it has no hand, but anything small and light could be added as a hand
- a second one may be added soon

*the arm has a specific name but i cant for the life of me remember what it is

with that in mind has anyone got any ideas for the project, because we're all rather overwhelmed with the choices
 
Last edited:
Remember, a good project wont simply be engineering a solution but will be trying to understand a scientific hypothesis. making original scientific contributions s difficult at the undergraduate level, but to get a First class kind or mark you will need to show good work towards such a goal. Furthermore, your project supervisor should have ideas of interesting research that can be conducted in a reasonable amount of time, I would go and talk to them.

However, the first step is to read a large a amount of background literature on the problem domains you are most interested in. Try to find out if their are any resolved findings, missing data, poorly understood mechanisms, limited support of a theory, etc. Since you will ave limited time you also have to make sure that the problem is realizable in the short amount of time you have.

My undergrad project in AI involved understanding environmental factors in the spatial co-evolution of cricket mating calls in spiking neural network through Cartesian spatially diverse GAs and Island models. I discovered supporting evidence for claims often stated in in biology papers about the genetic landscapes and spatial relatedness between cricket species. This could have formed a basis for my PhD but I changed directions somewhat.


I would say with your hardware available that their is a lot of recent progress in reinforcement learning which could make for an interesting but very challenging project, depending on how good your maths is.
 
AHarvey said:
what about simply making it so it mimics someone standing in front of it?

how simple or complicated a project are you looking for?
Click to expand...

depending on how helpful the kinect would be that could range from slightly too easy to about the right difficulty i would say.

to give you an example the previous tasks were to catch a ball thats rolled down a table towards it and another was to light a candle with another lit candle. i believe these were done without the kinect.
our lecturer has suggested making it learn how to throw a ball into a cup. he also suggested another program to pick the ball up with its other arm and automatically reset itself because of how long the learning process is likely to take with a standard neural net (if we decided to use one)




@D.P. thanks for the advice, i hadn't thought about looking for an area in need of exploration but now you have suggested it the idea sounds like a good one. i'll get reading :)
 
You must log in or register to reply here.
Back
Top Bottom