Bookmark and Share

Survey

Last post 05-22-2009 7:26 AM by gauna. 11 replies.

Page 1 of 1 (12 items)
Sort Posts: Previous Next

  • 05-22-2009 7:26 AM

    Survey

    Reply Contact

    hi, my friends.

    maybe 10 days waiting is too long for us. why not let us move our attention and talk some more interesting things. in passed one month, all of us do a very great job. now, let share our fruits. here is a simple survey. i hope it can be a good start.

    question 1. which game you prefer to, somu or mars? and why?

    2. which part do you think is most challenge in mars, and why?

    3. which part is your most satified in your solution, and which is most dissatified?

    4. do your have some ideas that you think is quite good, but you did not realize/code it? if yes, why?

      thank you for your attention!

     

  • 05-23-2009 12:02 AM In reply to

    Re: Survey

    Reply Contact

     

    1. Mars, because in the Mars, the robot has more sensors for me, and except the heat shield, everything is standard, so it can be a little bit easier.

    2. The part to control the arms, for we don't know how to do it and no hints or data like other parts.

    3. The part to find all the accurate location is my proud, but I got quite dissatisfied with my computer seems very slow, when it runs and I think it has something to do with my solution.

    4. I don't know what you mean.

    and ... Hopes more guys can attend in this survey.

     

    Clark Chan

  • 05-23-2009 12:36 AM In reply to

    Re: Survey

    Reply Contact

    chen xi:
    4. I don't know what you mean.

     

    i mean some good ideas you tried to "teach" you rover to do, but for some reason(maybe technical promble or limited by the time) you did not realize them in your final project. for me, i hope my rover can remeber where it has gone and draw a map by its memory. then use this kind data to do path planing. i have "teached" him how draw a map. but since this uncompleted data is quite difficult to use, in final, i just use it to help chooce local derection. 

    thank you! 

     

  • 05-23-2009 12:30 PM In reply to

    Re: Survey

    Reply Contact

    1,sumo, because it's first time I really use microsoft robot studio, everything is fun. 

    2,at first it's difficult to climb the slope, I try to lead it to a point near #2, only 2/5 possiblity that the rover would not roll down.Then use other algorithm and it's ok now

    3, I am glad that my solution is simple.

    4,For sumo, I used a new neural network model, training it to learn everything by itself,it need to run several times, but I destoryed the memory file.. I am busy these days and do not have time, so I just change 2 or 3 number of the sampled file and get 2:1, I beleive many people did this..change the number of wandering speed (like (300,300)->(300,-300))and picture processing area  (like 1/4->1) you can get 50% percentage to win.

  • 05-23-2009 3:23 PM In reply to

    Re: Survey

    Reply Contact

    1. I liked mars better, a lot more challenging, made it so much more interesting.

    2. This is tough, it's either when I was trying to look for the heatshield, or getting out of the crater, I guess getting out of the crater since the heatshield, after a while of thinking, was just my having the robot look for a big concentration of gray.

    3. I'm satisfied that I managed to make my Rover almost completely dynamic, that is, even when I changed the order of the waypoints it can still complete the mission. I'm particularly dissatisfied with how I managed to get the focus of the crater, it's not particularly elegant since it assumes there will be at least 3 points around the rim of crater, and also, I'm not satisfied with how it gets out of the crater, because I ended up having to resort to trial and error as opposed to hardcoding the path.

    4. If anything, rock avoidance. My rover does avoid waypoints that are neither its destination nor origin, so it kinda does avoid rocks, but I don't actually have it really avoid rocks. I mean, it can run them over, doesn't seem necessary.

    "Make everything as simple as possible, but not simpler" ~ Albert Einstein

  • 05-23-2009 9:38 PM In reply to

    Re: Survey

    Reply Contact

    thothling:
    2,at first it's difficult to climb the slope, I try to lead it to a point near #2, only 2/5 possiblity that the rover would not roll down.Then use other algorithm and it's ok now

     

    could you talk some more detail about you new algorithm and the exact possibility of it?

    thothling:
    4,For sumo, I used a new neural network model, training it to learn everything by itself

     

    what you want your sumo to learn? i mean the output of the network.

  • 05-24-2009 5:46 AM In reply to

    Re: Survey

    Reply Contact

     

    chen xi:
    4. I don't know what you mean.

    4. I want the robot go to #6 by going through the crater, for it can save a lot of time, but in most of cases, the rover will fall down to the bottom, and I believe there's a better way to control the rover, but time is limited so I didn't improve it.

    How about you lin?

    Clark Chan

  • 05-24-2009 8:24 AM In reply to

    Re: Survey

    Reply Contact

    i also go to #6 by going through the crater. however i think go around crater much more reasonable. i do this because i did not finish my path planning fuction as i have said. but i have a climbup fuction, it can help rover climb up safely. basic idea of this fuction come from classical algorithm "hill up". 

    thanks

  • 05-25-2009 6:46 AM In reply to

    Re: Survey

    Reply Contact

    1,It's just ..if the rover may fall over, try another way.

    2,input is image and other sensor information, learn how to act at that moment

     

  • 06-03-2009 7:42 PM In reply to

    Re: Survey

    Reply Contact

    invite all finalists to do this survey:)

    thank you

  • 06-03-2009 8:04 PM In reply to

    Re: Survey

    Reply Contact

    It's nice to have this feedback around here.

    3. For the path planning, I've used some O(N log N) implementation of Dijkstra's shortest path (with priority queues) so that it could go around the crater.

    The basic idea was to run some sobel algorithm (or alike) over the satellite top picture from the crater (which would be available in a real situation) and fit some circle around the crater (with some flood filling). And then I've tried to find the best ways to go down the crater, which were drawn using some Bresenham, when it really needed to go down.

    It was enough to give a reasonable path planning around the waypoints (it would automatically find the shortest paths in runtime and avoid going down the crater for the high cost edges of doing this in the graph). 

    As the graph was too large for what was required, I just filtered it to half or one quarter of the points, so that the intermediate points in the path planning were not too close.

    I've also heard that some used the A* to do this, I'd be curious to know what were their results.

    I was rather unsatisfied with my image processing solution because it was only colour based (it would probably never work in a real situation). And I've enjoyed zero.lin's idea of using Stereo there :)

    That's it... it would be nice to review other ideas of coding the solution.

    Kind regards,

    Daniel

  • 06-04-2009 7:08 PM In reply to

    • gauna
    • Not Ranked
    • Joined on 10-23-2006
    • Posts 15

    Re: Survey

    Reply Contact

    1. which game you prefer to, somu or mars? and why?

    mars, because it have more sensors

    2. which part do you think is most challenge in mars, and why?

    recognized the safe paths, because it is difficult :P

    3. which part is your most satified in your solution, and which is most dissatified?

    recognized how leave the crater with the cams and the most dissatified is the heatshield searching, because it is the only part where I use a littel hard code.

    4. do your have some ideas that you think is quite good, but you did not realize/code it? if yes, why?

    no
Page 1 of 1 (12 items)