Artificial Intelligence and Mobile Robots: Successes and Challenges

David KortenkampNASA Johnson Space Center

Metrica Inc./TRACLabsHouton TX 77058

kortenkamp@jsc.nasa.govhttp://www.traclabs.com/~korten

Outline

• Where we’ve been– Previous mobile robot competitions– Contributions to the state-of-the-art

• Where we are– Status checklist– Challenges to the AI community

• Where we are going– 2000 Mobile Robot Competition and Exhibition

The First AAAI Mobile Robot Competition (San Jose CA 1992)

• Large arena with fixed obstacles• Robot needed to find and visit 10 poles• Teams could mark poles in any way

they wanted• Contributions

– Sonar-based obstacle avoidance

• Assessment– Robots could move w/o hitting things– Robots could find large, marked objects

Second AAAI Mobile Robot Competition (DC, 1993)

• Office-like environment• Robots modifying environment• Contributions

– Three-layer architecture (Erann Gat, JPL)

• Assessment– Office navigation still unsolved– Start of real-world tasks

Third AAAI Mobile Robot Competition (Seattle, 1994)

• Office building environment is continued• Mobile manipulation is introduced• Contributions

– Successful multi-robot strategy (Georgia Tech)– Probabilistic office navigation (Stanford)

• Assessment– Office navigation more robust– Distributed multi-robots are fast

Fourth AAAI Mobile Robot Competition (Montreal, 1995)

• Office navigation is extended with human interaction

• Mobile manipulation with ability to distinguish objects (trash vs. recycle)

• Contributions– Color vision – Wheelchair applications

• Assessment– Non-trivial vision becomes possible

Fifth AAAI Mobile Robot Competition (Portland, 1996)

• Office navigation with occupancy detection• Mobile manipulation of moving objects• Alan Alda!• Contributions

– Fast color vision– POMDP navigation techniques

• Assessment– Office navigation is solved

Sixth AAAI Mobile Robot Competition (Providence, 1997)

• Real-world tasks (vacuuming, search)• Direct interaction with AAAI attendees (serving

hors d’oeuvres)• Contributions

– Multi-media interfaces– Entertaining robots

• Assessment– Human-robot interaction still young

Seventh AAAI Mobile Robot Competition (Madison WI, 1998)

• Robots move out of the arena• Hors d’oeuvres is continued• Sony robot “dogs” make appearance• Contributions

– AAAI Best Paper to Thrun et al for museum tour robot based on software displayed in previous competitions

• Assessment– Human-robot interaction making progress

Eighth AAAI Mobile Robot Competition (Orlando, 1999)

• No arena at all, robots roam through conference hall on scavenger hunts

• Hors d’oeuvres competition a big hit• Contributions

– Introduced a “challenge” competition

• Assessment– Continued improvement in hors

d’oeuvres serving

Y2K: Where are we?

• What is the current state-of-the-art?• What are the remaining challenges?

CMU and NASA take the Nomad mobile robot to Antarctica in January 2000 to search for meteorites

Mapping and Navigation• State-of-the-artüObstacle avoidanceüOffice building navigationüPublic navigation (with assistance)üMap construction (including 3D) using sonar and laser

• Challenges•Outdoor navigation (even with GPS)•Vision-based navigation

Robot Vision• State-of-the-artüColor vision for object recognitionüObstacle avoidance and cliff detectionüAutonomous highway drivingüActive stereo vision for tracking

• Challenges•Object segmentation•Robust landmark discovery and

recognition

Mobile Manipulation• State-of-the-artüStand-alone manipulators on mobile basesüUncoordinated base and arm motion

• Challenges•Coordinated base and arm motion•Multi-arm robots•Eye-hand coordination

Human-Robot Interaction• State-of-the-artüSimple tracking and gesture recognitionüFace recognition under ideal

circumstancesüCOTS voice recognition

• Challenges•Discourse management tied to task

context•Learning from humans•Adjustable autonomy

Control Architectures• State-of-the-artüLayered architectures to integrate reactive and

deliberative componentsüRobust execution of procedures

• Challenges•Architectures with learning “built-in”•Sharing components (“plug and play”)•Adjustable autonomy•Validation and verification

Multiple Robots• State-of-the-artüHomogeneous robotsüDivide-and-conquer tasks

• Challenges•Architectures for coordination of autonomous,

heterogeneous robots•“On-the-fly” teaming of robots•Distributed sensing•Tasks that require multiple robots

Challenges to the AI Community• Planning

– Planners that can “take advantage” of underlying robust execution systems

– Mixed-initiative planning with robots

• Learning– Fast, on-line learners that handle uncertainty– Improving off-line models from data

• Natural Language– Discourse tied to task contexts and agent

actions

Challenges (continued)• Knowledge Representation

– Representing perceptual information– Moving knowledge between continuous and

discrete (or symbolic) representations

• Applications– Apply known mobile robot results to other

domains that have similar characteristics, I.e., “immobots” (Williams & Nayak 1996)

– Use mobile robots!

The 2000 AAAI Mobile Robot Contest and Exhibition

• Goals– Foster the sharing of research ideas and

technology– Allow research groups to showcase their

achievements– Encourage students to enter the fields of

robots and AI– Increase awareness of the field

• Alan Schultz, Chair

Events• Contest

– Urban Search and Rescue– Hors d’oeuvres– Lisa Meeden, Event Chair

• Challenge– Tucker Balch, Event Chair

• Exhibition– Marc Bohlen and Vandi Verma, Event

Chairs

• Plus, Botball and Robot Building Lab

Urban Search and Rescue• Robots must enter fallen structure

and search for victims• Robots judged on

– Number of victims found– Relaying location of victims and

hazards– Communicating with victims– Innovative technology

• No teleoperation

Hors d’oeuvres• Robots serve hors d’oeuvres to conference

attendees at the reception • Robots judged subjectively by a panel for:

– Ability to serve food– Interaction with attendees– Manipulation– Sensing modes

Challenge• Robot must register and attend AAAI with

no a priori information • Judges will look for technical innovation• Components include:

– Start at front door– Navigate to registration (look at signs or ask

directions)– Register (stand in line, say name)– Attend talk

• Task pushes state-of-the-art

Exhibition• Demonstrations of robot technology that

does not fit within the contests • Scheduled demonstration times• Exhibition themes include:

– Learning by imitation (USC)– Learning by observation (Georgia Tech)– Multi-robot negotiation– Interaction with humans– Intelligent wheelchairs (U Texas)– Enabling technologies

Botball• High school robotics building and

programming competition – Over 150 teams participated

• Finals held at AAAI-2000– 46 high school teams competing– August 1 and 2 (finals at 3:00)

• Organized by David Miller of the KISS Institute for Practical Robotics and the University of Oklahoma

Sponsors• DARPA • Office of Naval Research• Naval Research Laboratory• AAAI