Slashdot Mirror
Denver Couple Unveils Homemade Service Robot
An anonymous reader writes "Jim & Louise Gunderson, owners of a Denver-based computer software tool development company, have finally unveiled their autonomous robot, Basil. Basil is completely home built, runs Linux with some instructions in Java, uses a sonar-based 'reification' logic system, and can go get you a beer or a pot of tea. Quoting: 'The plan is this: The Gundersons will ask Basil to go to the bar, request a couple of stouts from the bartender, and then, once they're placed on the titanium tray perched on his head, bring them back to his creators. They haven't told him how to do this — there's no set script in his processors that tells him to roll a certain distance southwest, speak a certain command, then come back. He'll have to figure it all out on his own, using a basic knowledge of bars and beers and so on, reasoning skills and an ability to understand certain parts of the world. When his sonars capture the image of a person, for example, he knows it's a person, not just a nameless object to be avoided. And he knows that, in this case, that person wants a beer.'"
At Denning we had a mobile robot security guard. It could roam a factory or warehouse looking for intruders. it had sonar, radar, and other things.
Notifying people of appointments, delivering small objects, and serving drinks is not only possible, it is probably the easiest set of tasks that you can do.
I have a project on-line that allows you to build a basic robot for $500. It has PWM motor control and basic tips on building the base. It uses a PS/2 mouse to do wheel encoders. (cheap) and using a USB A-D/D-A board to control stuff. (I won't give the URL for fear of slashdotting my server.)
So, my two points: 1) It is possible they are doing what they say they can do. 2) Its fairly trivial if you have the time to waste.
There's been much more progress in the last five years than most people realize, though. SLAM works now. Vision algorithms actually work. Low-cost inertial devices work. We're starting to see the payoff from the DARPA Grand Challenge, which gave robotics a serious and needed butt-kick.
In my humble opinion, the Darpa Grand Challenge, by offering a market to LIDAR makers, made vision-based SLAM a thing of the past and the under-budgeted : This beast has 64 laser telemeters on a rotating head. It gives a 100 000 3D points cloud of the environment 10 times per second. A working video slam seems to pale in comparison...
The Wise adapts himself to the world. The Fool adapts the world to himself. Therefore, all progress depends on the Fool.
In my humble opinion, the Darpa Grand Challenge, by offering a market to LIDAR makers, made vision-based SLAM a thing of the past and the under-budgeted.
That's what many of us with Grand Challenge entries once thought. Even Sebastian Thrun once thought that. But, in fact, the winning 2005 Stanford "Stanley" vehicle was running mostly on vision. Above 25MPH it was out-driving its LIDAR range. The vision system wasn't doing SLAM, though. It was comparing the road further ahead with the near road. If they "looked the same" (the machine learning system for making that judgment was the breakthrough) and the LIDARs profiled the near road as flat, then the vehicle could drive faster than it could stop within the LIDAR range.
For the Urban Challenge, LIDAR units were more useful, because the speeds were slower and the environment more cluttered. But see the current issue of IEEE Trans. on Robotics, the special issue on SLAM, to see how much progress has been made. It's useful to use a camera and a limited LIDAR together with a SLAM algorithm; the vision system brings in more data and the LIDAR has less ambiguity.
The Velodyne thing (which is a better-built version of the Team Dad spinning LIDARs of 2005) is a good device, but too big, too expensive, and has too much rotating machinery for a production product. I've met its designers and seen the thing. The next step will probably involve either flash LIDAR or MEMS mirrors. Eye-safe flash LIDAR is a reality, and if produced in volume, it wouldn't be that expensive. It's expensive now only because it needs custom ICs.
An affordable little non-scanning 3D LIDAR for indoor use would be useful. There's the Swiss Ranger, the first device that qualifies. This is a true 3D time of flight sensor with no moving parts and 176x144 pixels. It's been around for about five years as a custom research item, but it's now being sold as a product by Acroname for $7500. The price needs to drop by an order of magnitude or two, which is quite possible.
Even my human ears can tell the difference between some types of wall coverings based on ambient sound reflections.
Oh, there's a lot more potential for you than that. Humans actually be trained in echolocation. Blind people even pick it up, thinking they're using their face for it, and so it's been called "facial vision".
Information theory is life. The rest is just the KL divergence.
I saw a demonstration of Basil earlier this month at the event mentioned in the article, and the Gundersons explained some of the technology and what they are trying to accomplish.
There is nothing special about the sonar -- it's just a simple low-bitrate input scheme. The Gundersons are focusing on solving the problems of environment perception by focusing on a cognitive model instead of throwing horsepower at interpreting the input in fine detail, as computer vision or perhaps some sort of advanced sonar would. The robot manages an internal model of its environment, and compares the input to its expectations instead of continually trying to reconstruct a scene. Perhaps it distinguishes a chair from a person with clues (a chair doesn't move on its own, for instance).
Sorry about the confusion, but you are absolutely correct, the sonars are not classifying the material. In the lab there are several different types of chairs, and Basil has constructed sensor based models of most of them. The two main types are the four-legged wooden chairs and the wheeled office chairs. Basil uses the sonar patterns (Percepts in our terms) to distinguish between these based (primarily) on the different shape of the legs. The semantic tags attached to the percepts are 'wooden-chair' and 'short-wheeled-chair', so when the robot classified the chair in the video, it used the tag 'wooden-chair'
However, the robot is doing everything on its own, there is no joystick, no person wearing a headset just off camera. the robot has a complete planning/execution system, and uses a reification engine (for more details see our book "Robots, Reasoning, and Reification") to semantically tag what it senses.
Sorry about the anonymous posting, I'm waiting for slashdot to send me a password,
Jim Gunderson
However, when the robot is given a goal ("deliver tea to the conference-table-area") the mental model is used to generate a symbolic representation of the world-as-it-is, along with the representation of the world-as-it-is-desired. At this point, the tag "wooden-chair" is used to extract information from the semantic memory (an ontology of facts and behaviors) and the linkages in the ontology allow Basil to reason about chairs with respect to the current goal. So he knows that chairs are generally stationary - they stay put, as opposed to people who move on their own; he knows that wheeled-chairs can be pushed out of the way, but the wooden-chairs can't, and that people can be asked to move, but chairs can't.
So at this point the label begins to be less arbitrary since it is now embedded in a complex knowledge structure. If we gave the chair the label 'battleship' (and if we had information about battleships in the ontology), Basil would generate different behaviors with respect to the object.
The classification scheme is fairly simple - primarily because his sensor modalities are thin. He builds a set of representations (classic pattern based templates) and uses these for both recognition and preafference (projecting what the world should look like).
When he is learning a new object, he checks to see if the patterns are mutually exclusive or if two or more objects can be classified from the same sensor data. If there is no way to distinguish between two classes, he reports both as possibles. These get loaded into the mental model and as he gets more views of the object the winnows down the possibilities.
So he is using multiple time and space separated views and 'thing constancy' as a principle to help him classify. There is a whole lot more detail in our book.
Basil is designed to learn from his experiences. He maintains a complete episodic memory at present. The task for the next year is to enable him to analyze these memories and generate new sensor representations, to subdivide existing representations, and to add new facts to his semantic memory. The tools that we will use will be a mix of standard machine learning techniques along with a technique that Louise developed for environments where not all features are salient to the classification.
Jim Jim