tdaniel@hubcap.clemson.edu (tod daniels) (03/07/91)
I just read an article in the new Discover magazine about robots that are controlled by responses to a set of behaviors. Rather than operating with a real-world model, these robots use a set of primitives that compete for control. EX: If the robot encounters a wall, the AVIOD OBSTACLE behavior assumes control and guides the robot around the wall. Then the other behavior resumes control. Does anyone have any information on this method or can they recommend any literature on the subject?? Furthermore, what do you think about his ideas personally?? -- ----------------------------------------------------------------------- "It's so lonely when you don't even know yourself" -- RHCP tod daniels -- tdaniel@hubcap.clemson.edu
iaf@ely.cl.cam.ac.uk (Innes Ferguson) (03/12/91)
I haven't seen the Discover article you mention, but I'm guessing it refers to work done by Brooks et al. at MIT. If this is so, then as a good starting point I would recommend "A Robust Layered Control System for a Mobile Robot" (R.A. Brooks, IEEE J of RA, 2(1), 1986). As a followup I suggest reading Robotics and Autonomous Systems 6(1&2), 1990. The latter is a special issue on Designing Autonomous Agents (Pattie Maes, guest editor) and reviews work related/similar to behaviour-oriented robots (e.g. reactive and situated action agents, computational neuroethology, etc.). I would also recommend a couple of articles in Artificial Intelligence 47, 1991. The first is by Brooks - "Intelligence without representation" - and the second is by David Kirsh - "Today the earwig, tomorrow man?" Kirsh's article is a response to Brooks' claim that conceptual representation is unnecessary for designing agents capable of "intelligent" behaviour. IMHO, Brooks' ideas are important as he is suggesting a very different approach to creating artificial intelligence from the more traditional (symbolic, high-level, inference-driven) approach. However, I think his stance is somewhat extreme and I doubt very much whether his subsumption architecture can be scaled up to deal with such tasks as predicting other robots'/agents' intentions, learning, reconciling local and global constraints on a robot's own (multiple) goals, problem solving, language understanding, etc. _without_ resorting to explicit representations, inference (incl. default reasoning), or some run-time decision making procedure. I guess time will tell. Regards... Innes ============================================================================= Innes A. Ferguson Computer Laboratory, University of Cambridge, Pembroke St., Cambridge CB2 3QG, England. BITNET: iaf@cl.cam.ac.uk JANET: iaf@uk.ac.cam.cl Tel.: +44 223 334421 =============================================================================
bleck@ai.mit.edu (Olaf Bleck) (03/12/91)
|> If the robot encounters a wall, the AVIOD OBSTACLE behavior assumes control |> and guides the robot around the wall. Then the other behavior resumes |> control. |> |> Does anyone have any information on this method or can they recommend any |> literature on the subject?? Write or call MIT AI Lab Publications Office 545 Technology Square Rm 818 Cambridge, MA 02139 617-253-6773 Ask for a bibliography or papers by Rodney Brooks. |> Furthermore, what do you think about his ideas personally?? Well, I work for him... Enough said! Actually, this is a really touchy question in the AI community, particularly when it comes down to proving that this behavior based programming actually works and isn't just a hack. In fact, part of the problem is that there isn't really a universally acceptable mechanism set up to test it against. Rod's latest paper, "Integrated Systems Based on Behaviors", addresses some of these issues. Tear Here: ----------------------------------------------------------------- _________________________________________________________________ Olaf Bleck bleck@ai.mit.edu Research Scientist MIT Artificial Intelligence Lab Mobile Robotics Group