4

I am new to SLAM and autonomous driving.

Case:

Robot needs to start from point A and reach point B with maximum accuracy as well as precision for all the cases. The only object around it is the table (shaded box in the pic) and the ground it travels over. So, very less features. There aren't walls around for the conventional SLAM methods. The square signifies a reference object (a table for instance). The pic shows a top-view of the cases.
enter image description here

I can try all possible hardware to implement it (IMU, GPS, RGBD camera, etc), except 3D LiDAR like Velodyne. How can I help my robot get from A to B with accuracy and precision ?

Pe Dro
  • 2,651
  • 3
  • 24
  • 44
  • How points A and B are defined and in which format are they provided? What is accuracy and precision in your case and how do you compute it? – fdermishin Nov 08 '20 at 19:29
  • There is no specific answer to that. Because, I feel that it depends on the method I choose. For e.g. using graph SLAM will convert everything into 3D coordinates. I need accuracy of 5cm in each axes. I measure it as Cartesian distance from A and B. I am open to methods. – Pe Dro Nov 09 '20 at 03:23

1 Answers1

0

To me, a pad that emits infrared light seems like a good solution, assuming you're not dealing with many obstructions. Maybe have some downward facing photosensitive thing to locate it specifically in the center, and a few cameras around the perimeter. Then just "follow the light". Practically, you might need some sonar or something to keep it from bumping into table legs.

On the off chance one of the table legs occludes the pad completely, having it wander around until it sees the pad seems like it might be good enough, hard to say. If you have a hard time locating the center, you could get fancy with magnetic fields, patterns, etc. Hall effect sensors are really a magical thing.

sample 3d modle

Warlax56
  • 1,170
  • 5
  • 30
  • Adding something to the target (table) won't be possible in my case. – Pe Dro Nov 09 '20 at 05:06
  • @PeDro, do you have any way of locating your target? – Warlax56 Nov 09 '20 at 06:29
  • I can use Camera (RGB) or depth (but depth sensors aren't that good for outdoors). We tried using a segmentation algorithm but it required a particular portion of thr panel to be visible and was just fine with accuracy. – Pe Dro Nov 09 '20 at 07:00
  • Maybe using machine vision stickers in relevant locations would be a good solution. You'll have to get a bit creative with how to orient based off them. – Warlax56 Nov 09 '20 at 15:59