TL;DR
This paper presents a novel LQ-servo control strategy for two-wheeled self-balancing robots, enhanced with a reduced-order observer. This approach aims to achieve precise control with fewer sensors, reducing cost and complexity while maintaining robust stability and zero steady-state error, validated through simulations and hardware-in-loop testing.
What Problem Does This Paper Address?
Two-wheeled self-balancing robots (TWSBs), like Segways, are fascinating for personal transportation and recreation. However, their inherent instability and nonlinear dynamics make them challenging to control. While Linear Quadratic Regulator (LQR) offers good stability and optimality, it struggles with tracking problems due to steady-state errors and requires full state information, which can be expensive to obtain with numerous sensors or complex full-order observers (like Kalman filters). This paper tackles the problem of designing a robust and cost-effective controller for TWSBs that can achieve precise tracking without requiring all states to be directly measured.
Methods
The core of this paper is the combination of an LQ-servo controller with a reduced-order observer.
- LQ-Servo Controller: This controller is an extension of LQR that incorporates an integrator. This addition is crucial for eliminating steady-state errors, a common issue with standard LQR when dealing with setpoint tracking. The LQ-servo design aims to minimize a quadratic performance index, balancing control effort and system response.
- Reduced-Order Observer: Instead of a full-order observer that estimates all system states, a reduced-order observer is employed. This observer leverages the available sensor measurements (outputs) to estimate only the unmeasured states. This significantly reduces the number of sensors required, thereby lowering implementation costs and system complexity. The observer gain is designed using a pole-placement method to ensure stability and robustness against sensor noise.
- Hardware-in-Loop Simulation (HILS): To validate the proposed control strategy in a realistic, real-time environment, the authors utilize a Hardware-in-Loop Simulator (HILS). This allows them to test the control logic on an FPGA-based platform connected to a simulated robot, mimicking real-world conditions including noise and disturbances.
Datasets
Not discussed in the paper. The paper focuses on the mathematical modeling and simulation of a TWSB, using parameters from existing literature (Table 1 and Table 2).
Key Insights and Ideas
- Cost-Effective Control: The most significant insight is that by using a reduced-order observer, the need for expensive, high-precision sensors to measure all states can be drastically reduced. This makes advanced control strategies more accessible for practical TWSB applications.
- Zero Steady-State Error with Robustness: The LQ-servo controller, by integrating an error feedback loop, effectively eliminates steady-state errors for step responses. This is a critical improvement over standard LQR for tracking applications.
- Balancing Performance and Complexity: The paper demonstrates that it's possible to achieve high performance (minimizing overshoot and oscillation) and robustness without resorting to overly complex full-order observers or an excessive number of sensors.
- HILS Validation: The use of HILS is a strong point, showing that the proposed controller behaves well in a real-time, noisy environment, which is crucial for the practical deployment of such robots.
Key Results
The paper presents simulation results and HILS test outcomes:
- Simulation Results (PC-based): Figures 8 and 9 show that the proposed LQ-servo controller with a reduced-order observer effectively controls the position and angle of the TWSB. The settling time for the controlled angle and position states is approximately 8 seconds, meeting the specified requirements. The estimated velocity and angle rate closely follow the real states.
- HILS Test Results: Figure 12 shows the time responses of position and angle during HILS testing. While the position settling time is slightly slower than PC-based simulations without noise, the angle state remains stable even with the introduction of noise. Figures 13 and 14 demonstrate that the estimated velocity and angle rate from the observer closely track the real signals, particularly after 15 seconds for velocity.
Limitations and Future Work
- Sensor Reliability Assumption: The paper assumes that the outputs obtained from sensors are reliable and not significantly affected by noise. While the observer is designed to mitigate some noise effects, extreme noise levels or sensor failures are not explicitly addressed.
- Linearization: The control design is based on a linearized model of the TWSB. While effective for many operating points, the performance in highly nonlinear regimes might be limited.
- Future Work: The authors suggest that the proposed method can be further applied to reduce the cost of sensor installation and the complexity of high-order filter design for TWSBs, potentially leading to more cost-effective and performant robots compared to traditional LQR techniques. The HILS results also highlight the robustness of the angle state in the presence of noise, suggesting potential for further investigation into noise handling.
Summary updated