• Home
• Research
  • LAAS
  • SB-JPALS
  • AAR
  • GEAS
  • Laser Nav.
  • Leonard
  • UAV
  • iGPS
• Publications
• People
  • Current
  • Alumni
• About Us
• Useful Links
  • ION
  • GPS library
  • Trimble tutorial
  • GPS intro.
  • Stanford GPS Lab
  • GPS calendar
  • GPS eph FTP
  • Gps Alm FTP
• IIT Links
  • IIT
  • MMAE
  • MyIIT
  • WebMail
  • Storage

Leonard
The design and construction of Leonard, a prototype differential-drive AGV, was on many aspects a multidimensional, multidisciplinary task. While introducing an innovative analytical design process which combined mechanics, dynamics and controls, we built a GPS-based navigation algorithm, and performed the manufacturing phase. Our lack of past experience made it even more challenging and educationally enriching. In our lab, Leonard will serve as a mobile platform for the application and experimentation of GPS-based sensor technologies.

To start with, we agreed for a general vehicle architecture. The vehicle is made of two motors in the back, whose difference in angular velocity performs the steering through the two driving wheels. Two front floating casters provide stability. The DGPS necessitates a GPS receiver, a data link and their respective antennas on top of the vehicle’s body. Measurements of the motors’ angular velocities are brought by encoders. An embedded computation unit and the necessary interfaces makes the autonomous control possible. All those elements are enclosed in a hermetic aluminum frame.

Once the main features were in place, we used a three step iterative process to achieve high dynamic performance while taking into account the constraints imposed by the market supply. In each iteration, the dynamic model is updated according to the available devices and simulations are performed to predict the final performance. The process is repeated until a satisfying solution is found. This analytical method helps avoiding high-cost trial-and-error processes.

Thus, in a first place, a second order dynamic model for the vehicle and the motors has been derived and verified using three different methods: Newton’s equations, Lagrange’s and Kane’s methods for non-holonomic systems. Linearized along a straight line, the resulting fifth order state space representation has been used in detailed covariance analyses and direct simulations in order to identify and optimize critical parameters such as the location of the GPS antenna and the center of mass. It has also helped to measure the potential benefit of encoders and the importance of the sample time. Then, in parallel, the navigation code on-board Leonard has been built based on a tracking algorithm which drives the vehicle along one direction. Measurements for the closed-loop control of the system are obtained using CDGPS and encoders. The optimal corrections to the motors’ reference input are computed using a Linear Quadratic Regulator and a Kalman Filter updated in real time. Finally, although the manufacturing phase has revealed many unexpected hardware problems, the final product confirms the initial analyses and illustrates the benefit of pre-manufacturing covariance analysis-based design.

In future works, the construction of similar vehicles should make it possible to study the performance of distributed control for GPS-based systems. Also, other technologies such as INS or pseudolites may be implemented and experimented as augmentation sensors to the GPS.