Abstract
This work investigates the control system of a mobile robot, specifically focusing on trajectory control—ensuring a vehicle follows a pre-defined path without human intervention. The study analyzes various control architectures and proposes a functional architecture centered on optimal control theory applied to a mobile robot model in a fixed reference frame. A key contribution is the application of the Autonomous Trajectory Generating Servomechanism (ATGS) technique to improve controller robustness. The performance of the ATGS-enhanced controller is compared against initial designs through simulations and experimental data obtained from the "Ariel" robot, developed at the University of São Paulo.
1. Introduction
In the early 1990s, the field of mobile robotics sought to move beyond simple Automated Guided Vehicles (AGVs) toward autonomous systems capable of navigating arbitrary paths. A fundamental challenge in this transition was trajectory control: the ability of a robot to track a desired path (lines, arcs, or complex curves) despite initial positioning errors and hardware limitations. This paper presents the development of a trajectory controller for the "Ariel" robot, utilizing optimal control principles to achieve robust performance.
2. System Modeling and Linearization
The "Ariel" robot is a three-wheeled vehicle featuring two rear traction wheels and a front caster. To design an effective controller, the kinematic and dynamic equations of the robot were first derived. Given the inherent non-linearity of robot dynamics, the system was linearized around nominal trajectories—specifically straight lines and circular arcs—using Taylor series expansion. This linearization allows for the application of linear control techniques while maintaining accuracy within the operational envelope of the robot.
3. Optimal Control and the ATGS Technique
The core of the control strategy is the Linear Quadratic Regulator (LQR). By minimizing a quadratic performance index that balances tracking error against control effort (voltage/torque), the author derived optimal feedback gains via the Riccati equation.
A significant innovation presented is the use of the Autonomous Trajectory Generating Servomechanism (ATGS). Unlike traditional controllers that calculate error relative to a fixed global coordinate, ATGS employs a moving reference frame that slides along the intended trajectory. The error is defined as the distance to the closest point on the path. This approach renders the system invariant to global rotations and translations, significantly simplifying the control logic and preventing "panic" responses (motor saturation) when the robot is far from its target path.
4. Experimental Implementation
The proposed controller was implemented on the Ariel robot using an IBM PC XT equipped with an Intel 8088 processor (8MHz). The control software was written in C, operating with a sampling period of 15ms. This implementation demonstrated that sophisticated optimal control laws could be executed on the limited hardware of the era through careful discretization of continuous-time equations.
5. Results and Discussion
Validation was conducted through both simulations (using SIMNON and MATLAB) and real-world experiments. Key findings include:
- Robustness to Initial Errors: In tests with a 0.5-meter lateral offset, the ATGS-based controller successfully converged to the path without the oscillations or motor saturation typical of fixed-frame controllers.
- Path Versatility: The system successfully executed complex "Figure-8" trajectories by concatenating line and arc segments.
- Load Handling: The controller remained stable under varying loads (up to 100kg) and speeds (up to 0.82 m/s), though performance at higher speeds suggested the need for non-linear compensation.
6. Conclusion
This work successfully bridged the gap between theoretical optimal control and practical robotic application. By combining LQR with a moving reference frame, the author developed a controller that is both robust and computationally efficient. Future work identified the need for state observers (such as Kalman Filters) to mitigate sensor noise and the integration of obstacle avoidance layers within the broader control architecture.
Summary updated