Slashdot Mirror

← Back to Stories (view on slashdot.org)

Humanoid Robots For the Next DARPA Grand Challenge?

Posted by samzenpus on from the thank-you-very-much-mr-roboto dept.

HizookRobotics writes "The official announcement should be out very soon, but for now here's the unofficial, preliminary details based on notes from Dr. Gill Pratt's talk at DTRA Industry Day: The new Grand Challenge is for a humanoid robot (with a bias toward bipedal designs) that can be used in rough terrain and for industrial disasters. The robot will be required to maneuver into and drive an open-frame vehicle (eg. tractor), proceed to a building and dismount, ingress through a locked door using a key, traverse a 100 meter rubble-strewn hallway, climb a ladder, locate a leaking pipe and seal it by closing off a nearby valve, and then replace a faulty pump to resume normal operations — all semi-autonomously with just 'supervisory teleoperation.' It looks like there will be six hardware teams to develop new robots, and twelve software teams using a common platform."

14 of 53 comments (clear)

  1. Re:Bipedal humanoid by HizookRobotics · · Score: 3, Insightful

    This is a common misconception. In academia, "humanoid" will usually refer to a robot with _some_ humanoid features (eg. two arms). If you look at the "Humanoids" conference, you'll see that there are a healthy mix of legged and wheeled designs. So I thought specificity was good in this case...

  2. I'll be worried when... by Lord_of_the_nerf · · Score: 3, Funny

    ...the last part of the course is 'how to love' and 'question whether or not the unit has a soul'.

    1. Re:I'll be worried when... by MichaelSmith · · Score: 2

      I would be fine with them dreaming of electric sheep but I would draw the line at giving them memories of fictional parents.

      --
      http://michaelsmith.id.au
    2. Re:I'll be worried when... by dissy · · Score: 2

      I'll be worried when ... the last part of the course is 'how to love' and 'question whether or not the unit has a soul'.

      Installation of a soul is trivially easy, but the licensing costs are outrageous!

  3. That's ambitious by Animats · · Score: 4, Insightful

    This is very ambitious. Yet it's almost within reach, if enough money is thrown at the problem. A lot of money.

    First, this requires solving low-speed legged locomotion over difficult terrain. Walking over rubble and climbing a ladder will be very tough. But it's not totally out of reach. A machine which can walk on either two or four limbs would be an advantage over rubble. Ladder climbing is a four-limb problem, and it's been done at least once. Grabbing the rungs with all limbs will make it easier.

    Replacing a pump is an interesting problem. I don't think anyone has yet demonstrated complex part replacement in an unstructured environment. That's probably the least well developed part of the problem.

    Building a suitable machine will be expensive. Each Boston Dynamics robot has cost upwards of $20 million. Even the Willow Robotics machine is over $100K per unit, and it's just two arms on a wheeled base. This thing has to have roughly human dimensions and be self contained. Petman isn't self-contained; it needs external power.

    Yet, looking at the problem, I can see how to approach it. Modern control theory is good enough. Machine learning and vision processing are good enough. Simulators are good enough to allow debugging in simulation. Enough people know this stuff that the job can be staffed.

    1. Re:That's ambitious by Missing.Matter · · Score: 3, Insightful

      Machine learning and vision processing are good enough.

      I disagree on both counts. Machine learning is very good at specific problems. Current best techniques to train classifiers need millions of samples of ground truth training data, which is incredibly expensive to produce. Most classifiers like this make several assumptions about the training data, namely that it is representative of a stable underlying unknown distribution, which of course is hardly true. The best uses for machine learning techniques are for very specific applications where this assumption can be enforced, like optical character recognition. Machine learning has amazing applications, but it's hardly a field where I would say the results are "good enough," especially for most outdoor mobile robotics applications, like this DARPA project.

      Same goes for vision processing. The reason most robots today use LIDARs as their primary "vision" sensor is that image processing is very slow and not very good. With a 3d LIDAR you can do mapping, object detection and recognition, people detection, grasp planning, obstacle avoidance, etc. all very easily in polynomial time. Most of these tasks are computationally intractable using images, if they're possible at all (for example, night time operations).

    2. Re:That's ambitious by Animats · · Score: 3, Informative

      Same goes for vision processing. The reason most robots today use LIDARs as their primary "vision" sensor is that image processing is very slow and not very good. With a 3d LIDAR you can do mapping, object detection and recognition, people detection, grasp planning, obstacle avoidance, etc. all very easily in polynomial time. Most of these tasks are computationally intractable using images, if they're possible at all (for example, night time operations).

      10 years ago, that was true. Not any longer. Take a look at DARPA's current robot manipulation project. Watch a vision-guided robot with two three-fingered hands put a key in a lock, turn the key, and open a door. That's being done with stereo vision and force feedback. The gold-colored device with two lenses is a Point Grey stereo vision device.

      Much of early vision processing is done in custom hardware or GPUs now. It's not "computationally intractable". It's fairly expensive computationally, because the amount of work per pixel is high. But it's not exponential.

      When I ran a DARPA Grand Challenge team a decade ago, we relied on LIDAR too much, and vision not enough. The winning teams relied almost entirely on vision when they were going fast - the LIDARs didn't have the range or angular accuracy to read terrain out to the stopping distance. What they got from vision was basically "distant part of road looks like near part of road - OK to go fast if near part of road is good", obtained from a classifier system. So they could out-drive their LIDARs. We couldn't exceed 17mph.

    3. Re:That's ambitious by Missing.Matter · · Score: 2
      No it is true even today. Experiment's for DARPA's ARM project were conducted under bright lighting in ideal, static conditions. I've used the bumblebee cameras and other stereo vision systems, and they have limitations that inhibit the operation of many mobile robots including frame rate and maximum discernible range, (which is limited by the baseline of the camera), and most of all differences in lighting conditions.

      When I ran a DARPA Grand Challenge team a decade ago, we relied on LIDAR too much, and vision not enough. The winning teams relied almost entirely on vision when they were going fast - the LIDARs didn't have the range or angular accuracy to read terrain out to the stopping distance.

      Look at the more recent DARPA Urban challenges. Of the 6 teams that completed the challenge, all of them used the Velodyne 3D LIDAR as their primary vision sensor, in addition to several peripherial LIDARs. None of the teams who relied on vision primarily finished. Look at Google's Autonomous Prius. Its primary navigational sensor is the Velodyne. If vision were good enough, a car should be able to run on vision alone. After all, I do. But they can't.

      What they got from vision was basically "distant part of road looks like near part of road - OK to go fast if near part of road is good", obtained from a classifier system.

      Again, the assumptions of most classifiers are that the training data represent a stable underlying distribution. What happens if the road color changes? What happens if everything looks the same color, like in the desert or at night? With a LIDAR these things are no problem. With vision, a human can solve these problems. With vision, a computer cannot. Vision systems are not robust enough. Don't get me wrong, vision systems are important and can do many things a LIDAR cannot do, and in the future they will be indispensable, but they are simply not "good enough", especially for the task described in this DARPA challenge.

    4. Re:That's ambitious by Animats · · Score: 2

      Of the 6 teams that completed the challenge, all of them used the Velodyne 3D LIDAR as their primary vision sensor, in addition to several peripherial LIDARs. None of the teams who relied on vision primarily finished.

      True. The Urban Challenge was a slow-speed run, within LIDAR range. The latest Velodyne units see out to 70 meters. Note that you can't profile the road (to see potholes, etc.) out that far; the angle is too oblique. But you can see cars, pedestrians, telephone poles, etc.

      There are small short-range LIDAR devices, like the Swiss Ranger and photonic mixer devices, which have finally reached the Consumer Electronics Show level. For short ranges (a few meters), triangulation devices like the Kinect work. Longer range LIDARs without mechanical scanning do exist; Advanced Scientific Concepts makes them for DoD, at about $100K each. (The high price is not inherent in the technology; the manufacturer hand builds them in Santa Barbara, CA.)

      Again, the assumptions of most classifiers are that the training data represent a stable underlying distribution. What happens if the road color changes? What happens if everything looks the same color, like in the desert or at night?

      You slow down until the stopping distance is within the range of the near-range sensors. The Stanford vehicle's classifier was comparing the near road (out to 10-20m, and profiled by the LIDAR) image with the far road image. As the vehicle rolls forward, the far road becomes the near road, you find out if the far road was like the near road, and the classifier can be trained. After you've run on a reasonably uniform road for a while, you can speed up. All the vision system did in the 2005 Grand Challenge was to detect situations when the road was flat and smooth, allowing fast driving.

  4. Re:Why little to no teleoperation? by Mr.+Freeman · · Score: 3, Insightful

    You'd think so, yes. Of course, DARPA isn't so much about making useful things as it is about spurring research that can be used to later create useful things. I suspect that's why they are going with a competition that no one will be able to complete in the next 3 years, and certainly not in a reasonable time frame.

    This has happened before. It took two years before any competitor was able to even finish the off-road DARPA grand challenge, and that was just driving. A bi-pedal robot is orders of magnitude more complicated.

    I predict that they're going to have to start each section of the race separately. They'll put the bot in the tractor, start the race. Then they'll move the bot to the door, start it again, etc. No way in hell anyone is going to be able to do these tasks in sequence. Some they probably won't be able to do at all.

    --
    -1 disagree is not a modifier for a reason. -1 troll, flaimbait, redundant, overrated are NOT acceptable substitutes.
  5. Re:Humanoid Robot... by Sibko · · Score: 4, Interesting

    I am afraid not.

    A 60 ton Clan chassis has enough tonnage available, (assuming it moves 1/2/0, has no armour, and utilizes an XL engine with endosteel internals) but lacks the internal critical space necessary to actually fit all of that. (Staring at the design right now in solaris skunkwerks)

    Two LBX20's and 4 cStreak6's are pretty hefty firepower though. Even for a Clan heavy. Also, 12 tons of ammo? That is enough for 45 turns of non-stop firing, most matches do not even last 16!

    Now, I *could* build a custom mech from scratch that fits some of that, but I decided a better choice would be to take ye old venerable Mad Dog (Vulture for you Freebirths) chassis and rig it with the LBX autocannons, though I was worried it would not fit them AND the ammo. Thankfully, the Mad Dog's design principle follows the 'fuck armor, put more guns on it' school of thought.

    I yanked the dual large pulse lasers and LRM20's from the prime variant, and jacked the heatsinks down to the base 12 (It is in no danger of overheating with just these autocannons). This gave me 28 tons to work with. 24 tons for the two LBX20's, and then 4 tons for the ammo. It is not the loadout I would personally take into battle with a Mad Dog, considering its armour, but for a daring young Trueborn out to claim a bloodname, I do not think it is bad.

    A 5/8/0 movement curve (reaching ~80kph at a run), enough ammo to fire the autocannons for about 10 turns (1.7 minutes) before running dry, and just enough armour to keep you mostly intact as you close into range. At 1610 battle value (2222 BV with a 3/4 standard Clan Pilot) it is actually pretty dangerous for the price. I also made sure to mix the ammo types - 2 tons cluster and 2 tons slug so that you get your holepunching and sandblasting ability all in one. Additionally, I removed the upper arm actuators to clear out some of the internal crit space in the arms. So not only does the LBX20 assembly manage to fit in just the arm itself, but since it is entirely in the arm and the upper arm actuator has been removed, the arms themselves can completely flip around, making this design even more dangerous in a close quarters fight as you maneuver through a formation and fire *backwards* (Hopefully at the rear armour of your enemy.)

    The significant downside to this design is the staying power. Light on armour (But then, all Mad Dog's are) and light on ammo. Once you run dry, there are no backup energy weapons, so you will be left adopting dezgra freebirth tactics like kicking your opponent to death. (*spit*)

    Additionally, the design lacks anything to absorb hits in the torso. Any breach of the armour that scores a critical hit is going to either blow out an ammo bin and cripple the mech, or damage the fusion reactor. (Granted, a damaged fusion reactor on this design is not so bad - thanks to the low heat output of the cannons. Even with half the reactor shielding gone, this mech is still going to run cool.)

  6. Couldn't they use a pair of robots? by Snospar · · Score: 3, Funny

    Say one, perhaps gold in colour with a slightly annoying, camp voice, that's humanoid and another that's really the brains of the operation in a tri-wheeled body that communicates with a series of clicks & beeps (it's also got a useful rotary arm for shutting off that valve).

    It would make for a more entertaining video at the end of the competition.

    --
    Moore's law is not a law. Theory, yes; Predictable trend, certainly; Law, no.
  7. The case for Humanoid robots by scharkalvin · · Score: 2

    The advantage of a humanoid robot, especially a bipedal one is that it can make use of tools and vehicles designed to be used by humans. Such a robot can perform tasks too dangerous for a human that could be done by a human. More specialized robots in non human form can be designed especially for high risk jobs, such as clean up of a nuclear disaster. The Japanese have been specializing in humanoid robots for some time as they have special needs. With an aging and shrinking population they see the need to develop service robots to take care of the ill and elderly. A proposed 'first step' was Project Atom to develop a humanoid robot with the Intelligence level of a 9 year old child. (The name for the project comes from Tezuka Osmau's Tetsuwan Atom manga, better known outside of Japan as Astro Boy).

  8. Re:Bipedal humanoid by mcgrew · · Score: 2

    A double amputee isn't human? A kangaroo is humanoid?

    --
    Free Martian Whores!