ControlShell: A Real-Time Software Framework

Principle Investigators

Robert H. Cannon Jr., Professor, AA Dept., (cannon@sun-valley.stanford.edu)
Jean-Claude Latombe, Professor, CS Dept.,( latombe@cs.stanford.edu)
Stan Schneider, RTI, ( stan@sun-valley.stanford.edu)

Project Objectives

Forge an innovative, powerful real-time software framework and development environment. The framework will be shaped by the simultaneous development of cogent machine capabilities: experimental on-line motion planning and control in a dynamic environment and deft coordination of multiple interacting mechanisms and sensors.

Approach

The research will be build on ControlShell, originally developed at the Stanford Aerospace Robotics Laboratory and now marketed by Real-Time Innovations, Inc. The research will add fundamental new capabilities, including network-extensible data flow control to allow scalable support for distributed systems and a graphical environment to enable rapid prototyping and increased productivity. The concurrent development of this tool and experimental robotic applications is aimed at ensuring high-quality architectural design and producing reusable components.

New Ideas

On-line distributed planning.
Component-based framework for real-time software.
Network extensible data-flow architecture for planning and control.

Impact

Our work in distributed planning will provide machines with on-line motion planning autonomy. This will incorporate--for the first time--a capability that requires provably complex computations into a real-time system . This result should be generically applicable to solving computationally "hard" problems on-line in a real-time framework.
Our component-based approach to real-time software development will revolutionize the generation, maintenance and sharing of real-time software. Our approach of concurrent development and operational testing of both system and application software will insure a high-quality architectural design, and embed fundamental new contributions to the science of intelligent machine control into reusable software components.
Our highly-scalable distributed data-flow architecture will enable new, previously unachievable levels of vertical integration of complex control and planning systems. This will permit high-level task directions to be integrated with on-line motion planning and deft control into complete systems capable of interaction with a dynamic, physical world.

Subtopics

ControlShell Framework development
Distributed Control Architecture and Interfaces
On-Line Computation Distribution Architecture

Applications and Technology Transfers

Summary of the Main Result Obtained In the Past Year:

Have completed the redesign of the two cooperative robot arms.
Have experimented planning distribution along several axes we identified.
Have developed a general manipulation planner for multiple robot manipulators in 3D environments.
Have developed an online manipulation planner suitable for our real-time control architecture.

Summary of the Main Result Obtained In the Tenure of This Project:

Have completed the design of the initial interface between planning, control, and world modeling for the multi-robot manipulation demonstration.
Have developed a linear-time algorithm to compute a time-parameterized path of a robot given a list of via-points. This algorithm has been implemented as a ControlShell component, and is in the process of being proven experimentally on the robotic hardware.
Have designed and implemented a performant polynomial-time landmark-based motion planner for mobile robot navigation with uncertainty.

Acknowledgment:

This research is funded by Darpa/Navy Contract No. N00014-92-J-1809

Please send questions and comments to li@flamingo.stanford.edu