1
$\begingroup$

Will vision based system eventually prevail rangefinders based systems given that the most successful autonomous agents in the world like animals use primarily visions to navigate?

$\endgroup$
  • $\begingroup$ I'm not sure of the downvotes. this is fairly interesting, and we all know the next gen roboticists are biologists anyway. $\endgroup$ – Josh Vander Hook Feb 15 at 6:33
  • $\begingroup$ Some animals have evolved rangefinding that doesn't use vision. For example both Bats and Dolphins use echolocation. Some fish and eels use spatial electric field strength and direction. $\endgroup$ – DavidJ Feb 15 at 16:24
7
$\begingroup$

Animals and robots both need to understand something about the 3D structure of the world in order to thrive. Because it's so important, animals have evolved a huge number of strategies to estimate depth based on camera-like projective sensors (eyes). Many make use of binocular disparity -- the fact that the distance between the same scene point in two different views is inversely proportional to depth. A mantis moves its head sideways to do this, a dragonfly makes very precise forward motion, birds move their heads forward and backwards, we use the fact that our eyes are widely spaced on our heads and have overlapping fields of view. Stereo vision is a computer vision technique that mimics this.

However disparity is not the only trick we use, there are lots of others including proprioceptive signals from the muscles in our eyes (focus and vergence), observing how things in the world move as we move (optical flow), our knowledge about the world (the tree looks small, we know trees are big, therefore it must be far away), haziness, texture density, occlusion (thing A blocks thing B, therefore A must be closer). Each technique (or cue) is imperfect, but mostly in a different way, and each operates across a different range of distances. Our brain stitches (fuses) all that together to give a result that is mostly pretty robust.

Checkout the wikipedia page for more details. I like this (old) article

How the eye measures reality and virtual reality. JAMES E. CUTTING, Cornell University, Ithaca, New York. Behavior Research Methods, Instruments, & Computers 1997, 29 (1), 27-36

which gives a good insight into this issue.

There are now deep-learnt networks now that can take a monocular (single) image and output estimated depth. We can never know for sure but my theory is that they are encoding some of the cues that we use for depth estimation such as texture and apparent size.

$\endgroup$
3
$\begingroup$

Vision is actually one kind of rangefinder, although passive. Earlier camera use parallax to find the distance and focus.

Nowadays, active rangefinders are developed such as ultrasonic sensors, lidar sensors, which emit a signal and checks the returned time, the time of flight. This is more accurate and could adapt to low light conditions.

$\endgroup$
1
$\begingroup$

Short answer: They don't need. Cameras can be used as range finders, and an easy example is the Xbox Kinect.

Computer vision is an area that is being studied a lot. You can find a lot on authonomous cars, and the problem of SLAM.

$\endgroup$
  • 1
    $\begingroup$ The Kinect is not just a camera, it also has an infrared pattern projector which is crucial to its ability to measure 3D structure. $\endgroup$ – Peter Corke Jan 20 at 23:40
  • $\begingroup$ that still is a camera @PeterCorke $\endgroup$ – BlueDi Jan 24 at 15:20
  • 1
    $\begingroup$ @BlueDi that's quite wrong. A lidar is "still a camera" if you want to be glib about it, since it does sense light. Both LIDARs and Kinect project light and sense the response from the world. whereas biological eyes and SFM e.g. Stereo is passive. $\endgroup$ – Josh Vander Hook Feb 15 at 6:26
  • $\begingroup$ @JoshVanderHook Yes, just on the capture part. I agree on what you say about it using LIDAR, and that is why I referred Kinect. It is another way to capture an image of the world (instead of the usual camera) that current robots use. Maybe the correct way to put it is that "Rangefinders can be used as cameras". $\endgroup$ – BlueDi Feb 17 at 16:45
  • $\begingroup$ closer. I think "Cameras can be used as rangefinders" is right "the kinect is a camera used as a rangefinder" is kind of right but hides important details. but in light of the question of active versus passive (esp biologic) sensors, the kinect analogy just seemed wrong. $\endgroup$ – Josh Vander Hook Feb 17 at 17:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.