SBIR 95-1 SOLICITATION
PROJECT SUMMARY
Proposal Number:
Project Title:
ROBOTIC ONLINE LEARNING OF PERCEPTUAL MODELS FOR NAVIGATION AND GRASPING
Technical Abstract:
A crew helper robot must be able to observe new places and objects, construct
models automatically, and then use those models to enact instructions
requiring autonomous navigation and grasping. The proposed innovation is a
software module which: can encode the shape of any 3-D place or object to
limited or extreme accuracy, as desired; can learn shapes incrementally
from sensor data; supports sensor-based navigation and real-time object
recognition and tracking; is fully explicit so that represented volumes
and points can be identified, thus facilitating Artificial Intelligence
reasoning about the represented volumes; and supports route/trajectory
planning in 2-D, 3-D, and on orbit. Also solved is how the robot can
retrieve the newly learned place and object models by means of newly
learned names. Therefore, our representation is the essential glue between
a) sensor data obtained online, b) sensor-based robotic task performance,
c) Artificial Intelligence spatial reasoning, and d) Natural Language
instructions, even when the instructions involve places and objects not
known to the robot's programmer. NASA requires such a capability for robots
serving as astronaut assistants and as instructible maintainers of spacecraft
and lunar/planetary equipment.
Potential Commercial Applications:
Vendors of autonomous mobile robots for cleaning, sentry duty, mail delivery,
mowing, etc need a means of simplifying robot installation despite the
variability of floor plans and lawn areas. The proposed representation of
space will allow a robot to learn a floor/yard plan (in stages, if desired),
navigate robustly using learned landmarks, and obey instructions containing
user-selected names for the newly learned areas.
We are also considering applying the object recognition technology proposed
herein to parts sorting for industrial automation.
Name and Address of Offeror:
Robotics Research Harvesting, 166 Springdale Way, Redwood City, CA 94062
Small Business Innovation Research Program (SBIR) &
Small Business Technology Transfer Program (STTR) Programs
Electronic Management System (EMS)
National Aeronautics and Space Administration (NASA)
The SBIR/STTR EMS site is maintained by Steve
Hu, Hughes STX.
For comments and questions, contact
webmaster@sbir.hq.nasa.gov.
Updated: Feb. 7, 1996