Faculty:

We are developing a software framework that supports design of reconfigurable real-time software for embedded control systems [1]. We developed the notion of port-based objects (PBOs) to decompose, integrate, and reconfigure both control modules and device drivers in a component-based manner. Reconfigurable software has many advantages over conventional software design methodologies, including rapid development, changing software on the fly, supporting evolutionary design, allowing for remote upgrades and maintenance, and improving reliability through automated analysis [2].

This project, in collaboration with The Draper Laboratory (http://www.draper.com) in Cambridge, Massachusetts, is a first step towards applying the software model to a flight control system. Draper has developed an avionics processor and an unmanned aerial vehicle, whose development infrastructure we can use to perform our experiments. We would validate the stability of our software models, communication algorithms, and real-time scheduling, for use in their aerial vehicles.

A flight-control system represents one of the most difficult applications for embedded software, due to the catastrophic consequences of even the smallest errors. While we have successfully demonstrated our PBO software models in a variety of robotic projects, the robustness of the models and stability of the algorithms are still questionable, since they have not been applied to applications as critical as a flight-control systems. An autonomous helicopter, which in itself is a class of robot, needs dynamic reconfigurability for its new generation of advanced flight control systems. However, since our software has not been tested in such a critical environment, it is not proven safe to use.

The Draper Laboratory has created an infrastructure for developing flight control system software that first starts with development and simulation of the control algorithms in a simulation environment hosted on a graphics workstation, then moves to hardware-in-the-loop (HWIL) development and testing phase, and concludes with a fully embedded system. In the HWIL phase, the control algorithms are ported to the target avionics processor that executes in lock step with a simulation processor that provides sensor inputs and receives effector outputs. The target and simulation processors communicate over a VMEbus. Timing tests are done in the HWIL environment.

The simulation and HWIL environments for Draper's Small Autonomous Aerial Vehicle (DSAAV) will provide the basis for testing the stability and robustness of our reconfigurable software. The DSAAV shown above is a small unmanned helicopter, that was the winner of the 1996 International Aerial Robotics Competition. During the competition, it set a new record of seven autonomous flights, during which it performed real-time image recognition and object locating. This past year, the vehicle was demonstrated to the army at Fort Knox, Kentucky and Fort Benning, Georgia.

The DSAAV is relatively small, using a COTS small industrial helicopter with a total weight of 35 pounds and a 6-foot rotor diameter. It includes a 100 MHz 80486-based embedded computer, GPS, CCD camera, image mapping unit, sonar altimeter, and a magnetic compass. The software features advanced guidance, navigation, and control algorithms, a remote human systems interface, autonomous mission planning, and image-augmented navigation.

The software for the DSAAV is already robust, as proven by its successful demonstrations. However, there are several problems with the current software implementation that limit Draper's ability to change the control algorithms. Small changes require significant development time, as the software has not been designed for evolution. This is especially problematic in the newest generation of adaptive flight control algorithms, that require changing control algorithms in mid-flight. Real-time scheduling is performed using a fixed-priority scheduling algorithm, that does not make the most efficient usage of the available processor's power. Since power consumption and weight constrain the on-board processor, it is not possible to use faster processors, yet there is the desire to add functionality. Monitoring of the system's timing parameters are done manually and not very accurately. The current software models do not allow for automated monitoring.

The reconfigurable software that we will use can address all of the above issues. In particular, we developed the port-based object model to support software evolution and allow dynamic reconfiguration in real-time [1], and demonstrated its use for dynamically changing the software algorithms on robot manipulators to enable impact control [3]. The framework was designed with built-in timing error detection, handling, and monitoring, as we describe in [4] and [5]. The flight-control system is a more extreme application than the robotic manipulator, hence it serves as a good follow-up challenge to the software's stability.

[1] D.B. Stewart, R.A. Volpe, and P.K. Khosla, "Design of dynamically reconfigurable real-time software using port-based objects," IEEE Trans. on Software Engineering, v.23, n.12, Dec. 1997.
[2] D. B. Stewart and G. Arora, "Dynamically Reconfigurable Embedded Systems - Does it Make Sense?" in Proc of Real Time Application Workshop, Montreal, Canada, October 1996.
[3] R. Volpe and P. Khosla, "A Theoretical and Experimental Investigation of Explicit Force Control Strategies for Manipulators," IEEE Transactions on Automatic Control, v.38 n.11, November 1993.
[4] D. B. Stewart and P. K. Khosla, "Timing error detection and handling in real-time systems," Communications of the ACM, v.40, n.1, pp. 87-93, January 1997.
[5] D. B. Stewart and P. K. Khosla, "Policy-Independent RTOS mechanisms for timing error detection, handling, and monitoring," in Proc. of IEEE High Assurance Systems Engineering Workshop , Niagara, Ontario, Canada, Oct. 1996.