-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Yes for the most part. Once it bounces it is very difficult to backtrack and determine what path it took. So most LIDAR systems filter out implausibly weak LIDAR pulses or restrict the maximum wait time (e.g. maximum distance). Of course, extremely reflective objects can fool such algorithms. If you're driving between two giant angled mirrors, you will probably be in for a surprise. More practical examples would be chrome trim, speed signs, etc etc etc. It seems like in the Tesla community everyone is aware that overhead speed signs are bad for radar, but few realize that it's not great for LIDAR either. Both sets of sensors have their strengths and weaknesses.
To be fair, the best position for the radar is probably also elevated above the roofline and on a spinning turret. Kind of like airport radar. Except you look like a moron driving around with one of those, so that's something that will inevitably change once LIDAR exits the prototyping stage.
EDIT: To point out the above: Radar could potentially work well non-line-of-sight, because it can measure distance by phase shift of a frequency-modulated signal in addition to time of flight, and measure speed by doppler shift without having to glue multiple point-cloud readings together. As a result, you can gain a lot of information any time your radar waves reach another car and then back to you, regardless of what path it took.
Looks like a test mule. We've seen such things for years. I remember one festooned with cameras maybe 1 to 2 years ago. You can't read anything into this. It's like reading tea leaves in a cup of coffee.
thegruf
Active Member
This is somthing I have wondered for a while. Would it theortically be possible for the radar to bounce off the vehicle in front and pick up traffic in the adjacent lane beside or even behind you? presumably scatter would be a major issue but theoretically it might be possible to get enough concentration of a signal to indicate something, especially as you would be looking at a narrow range of angles. Somewhat akin to bouncing the radar under the vehicle in front. Can't see it being reliable enough for full indication to the driver but might be able to act as an additional "confidence" input to the probability of a vehicle next to you.
Yeah, radar can bounce off stationary objects and then continue to bounce off other objects. Being able to process that data is a bit complicated and quite interesting of a problem, and I think that's a relatively unexplored area for automotive systems (which goes back to what Elon was saying about advanced signal processing of RADAR).
But yeah, bouncing off a stationary object introduces a doppler shift equal to your speed. So if you're going 70, and the car in front of you is going 60mph, you will see a lot of radar signatures with a -70mph doppler shift (everything stationary), and you'll also see a strong radar signature for -10mph (the slower car in front of you). But you'll also see some other combinations, like the -60mph reflections bouncing off the car in front then to you, that could either add or subtract from the true speed.
Being able to process all of that into a coherent interpretation is definitely not a simple problem, but it's highly rewarding. That's what allows radar to potentially be superhuman in its ability to sense stuff that humans can't. I think seeing 2 cars ahead is a great initial example of this ability, but it definitely don't stop there!
Gizmotoy
Active Member
Indeed. It is possible to hide them better than most of the prototype self-driving systems. Our other vehicle has a (limited) LIDAR system that lives behind a piece of transparent plastic with a line painted on it to look like the rest of the grille. If I point it out it you'd find it obvious, but at a glance you might not notice. When paired with one or more radar, and one or more cameras, the complete set makes a pretty powerful object detection suite.
The fancy units like the Velodyne catalog are impressive, but you don't need to have a big spinning drum on your roof to get some benefit from LIDAR.
Despite Musk's earlier comments, I wouldn't be surprised if Tesla ships LIDAR in its Autopilot sensor suite over the next few years. Especially if the price continues to drop like it has been, and there's at least one company with an upcoming solid-state LIDAR puck that is reported to produce point clouds similar to the Velodyne units without spinning.
Exciting times.
Not impossible, but difficult and very sensitive to manufacturing tolerances, at least if you're planning to do electronic beam steering like the radar is using (and if you aren't then the blended in spots have to rotate...)
I don't know how much you know about phased array technology, but basically it steers the beam of energy by timing the emission of individual omnidirectional pulses - which have to be offset by a fraction of the wavelength to be effective in steering the beam.
Thus, the timing gets trickier the higher the frequency - for the LIDAR frequency above you'd either need to hold the exact length of each fiber optic fiber from the source to the emitter head to a precision a few dozen nanometers or to somehow calibrate the emitter head to know what the different lengths are so it can adjust the source timing to match.
On reflection, if the emitter head has the precision to manage the beam steering in the first place, you can probably do it with fiber optics and advanced processing - you just need to come up with a method for doing the calibration mentioned above (fire each element individually in a darkened room with a single reflector at a known distance, use timing to infer a fiber length?) and then use the fiber map to offset the steering phase timing.
thegruf
Active Member
thanks for the discussion
no specialist for sure, but for sure I get there would have to be a calibration process.
The line of thought was more along the lines that not many of us (currently ) would be enthralled by the idea of a lidar unit on top of our cars, and the prices have a long way to go to get terahertz equipment to commodity pricing, so would a semi internal unit with "fibre optic ducting within the vehicle skin be feasible
no specialist for sure, but for sure I get there would have to be a calibration process.
The line of thought was more along the lines that not many of us (currently ) would be enthralled by the idea of a lidar unit on top of our cars, and the prices have a long way to go to get terahertz equipment to commodity pricing, so would a semi internal unit with "fibre optic ducting within the vehicle skin be feasible
The spinning dome surely does wonders for aerodynamics too.
Initially I was thinking it was unrealistic, but as you saw I more or less reversed myself while trying to explain the situation. If you can build an effective inexpensive phased array LIDAR, then I think you can build it with a fiber optic remote transmitter with a little more computing effort.
I'm not sure how much that helps you overall, though - the emitter still needs to be something close to flat and perpendicular to the direction you want the signal to go (maximum practical steering is usually around 60 degrees away from axis, and you lose signal strength as you move off of the nominal axis.) The size of the emitter/receiver has implications in system sensitivity/power as well.
Maybe the most practical approach would be to put them into the headlight assembly and tail lights and one or both sets of side markers, providing the cost comes down enough.
Similar threads
