Rapid World Modeling

Description

Knowledge of the environment is a critical component of robotics automation. A robot must be aware of its environment to interact with it, for example, to avoid collisions while moving or to pick up an object. We call this knowledge the "world model".

Our world models are computerized descriptions of real world environments. This mainly consists of three dimensional (3D) surface geometry, which is used for operator display and control of our robotics systems. Additional information about the environment can also be stored with the model.

The Problem

Creating the world model takes too long. CAD modeling of a workspace is a manual, time consuming operation, usually taking hours or even days per object. Using traditional CAD tools can take days or weeks to create an adequate model of a workspace. Even after good models are created, they must be accurately registered into the robot's frame of reference to be useful. Add to this the problem that many objects, such as rubble or dented materials, are difficult or impossible to model with traditional CAD software.

The Solution

Using a Sandia developed sensor system, we can create or update a world model almost instantaneously from sensor data.

Our system surface geometry of the area to be modeled from data provided by 3D range sensors. Multiple scans are fused together, allowing an entire workspace to be modeled. We have created models of large spaces (40 X 60 X 20 foot warehouse) to small parts (2 X 1 X 1/2 inch).

We can easily accommodate data from various range sensors. In addition to our in-house structured lighting system, we have created models from Mechanical Technology Inc.'s structured lighting system, laser radar systems from Perceptron and Coleman Research, and several experimental systems.

Sandia National Laboratories is on the leading edge in rapid world modeling, providing a complete solution for bringing structure to an unstructured environment. A full graphical user interface integrates the sensor setup, data collection, editing, and surfacing of 3D range data. These models can be immediately incorporated into graphical programming and automated path planning systems. Various data reduction techniques have been incorporated to balance the size and geometric complexity of the world model.

Current work is focusing on texture mapping and advanced data reduction by fitting pre-made object models into the range data. Texture mapping allows us to map video images (or other data, such as heat radiation) onto the surface maps in near real time. Object fitting allows a user to place existing traditional CAD object models correctly into the world models' frame of reference. These techniques will enable us to present model and sensor data, from multiple sources, in a common framework.

Features

• Create models in a fraction of the time of traditional methods (within days, or minutes)
• Create models of complex surfaces previously impossible to model
• Complete solution - sensor, controls, model creation
• Uses only commercially available, standard hardware
• Easy to reconfigure for large or small areas, low or high resolution
• Enables robots to work in a previously unstructured environment
• Sensor fusion
• Interactive 3D visualization.

Collaboration

Our focus is world modeling of real-world environments, primarily to enable automated robotics to operate in an unstructured or changing environment. We work with other labs, universities, and industry to provide solutions to site characterization, automated handling, waste cleanup, and other environmental problems. The usefulness of world modeling goes well beyond our current emphasis. We are pursuing collaborative research in these and other applications.

Contact:
Charles Q. Little
(505) 284-3151
email: cqlittl@sandia.gov