TL;DR
This paper presents a novel control method for quadruped robots to achieve a "flying trot" gait, enabling faster and more agile locomotion. By combining offline trajectory planning with real-time balance control, the system mimics animal running dynamics, improving robustness and reducing impact forces through active compliance and a new posture control strategy.
What Problem Does This Paper Address?
Quadruped robots are increasingly being used in complex, unstructured environments due to their adaptability. A key trend in their development is improving movement speed. While various gaits exist, the "flying trot" offers higher speeds by incorporating flight phases. However, achieving stable and robust flying trot locomotion is challenging. Existing methods often combine vertical flight control with specific hardware (like pneumatic systems), limiting their universality, or lack detailed explanations for critical control aspects like foot lift-off and descent. This paper aims to address these challenges by proposing a unified control framework that integrates trajectory planning with real-time adjustments for stable, high-speed flying trot.
Methods
The authors propose a multi-faceted approach:
- Offline Trajectory Planning:
- SLIP Model: The Spring-Loaded Inverted Pendulum (SLIP) model is used to analyze the vertical dynamics of running animals. This informs the planning of the robot's Center of Mass (CoM) trajectory.
- Foot Trajectory Planning: Based on the planned CoM trajectory, the foot trajectories are planned in both the vertical (z-axis) and horizontal (x-y axes) planes. This includes predicting touchdown times for proactive landing buffer actions and controlling forward and lateral speeds.
- Directional Control: A rotation matrix is employed to adjust foot trajectories for changing the robot's heading, allowing for maneuvers and obstacle avoidance.
- Real-time Control:
- Posture Control: A novel position-based posture control method is introduced, utilizing virtual torques to maintain stability. This method focuses on controlling the support leg's action.
- Active Compliance Control: A leg virtual model control is used to further reduce impact forces during foot-ground interaction. This involves a spring-damping system for the legs, with parameters chosen to synchronize with the gait cycle.
- CoM Regulation: A kinematic-based CoM observation and regulation method is presented to enhance locomotion stability by ensuring the CoM projection stays within the support line.
- Parameter Selection: The paper details how various control parameters (step frequency, forward speed, foot positions, etc.) are determined, often by referencing animal locomotion data or through experimental analysis.
Datasets
Not discussed in the paper. The experiments utilize a quadruped robot named "Billy" and a Webots simulation environment.
Key Insights and Ideas
- SLIP Model for Planning: Leveraging the SLIP model for offline trajectory planning provides a biologically inspired and effective way to manage the complex vertical dynamics of running.
- Predictive Landing Buffering: Planning touchdown times in advance allows for proactive buffering, significantly reducing impact forces and improving robustness.
- Position-Based Posture Control: Shifting posture control to a position-based approach, rather than solely relying on speed adjustments, leads to smoother and more stable gait.
- Virtual Model for Compliance: Using a virtual model for leg compliance, with parameters tuned for resonance with the gait cycle, effectively enhances impact absorption and disturbance rejection.
- CoM Regulation for Stability: Actively regulating the CoM's projection onto the ground is crucial for maintaining stability, especially during dynamic maneuvers like the flying trot.
Key Results
- Simulation: The simulation in Webots demonstrated that the proposed control method can drive the robot to run with a flying trot gait, quickly stabilizing from external disturbances and achieving high speeds. The robot could track desired forward speeds with a deviation within ±0.2 m/s.
- Experiment: Experimental results with the Billy robot showed stable running with the flying trot gait, with no significant posture fluctuations. The average speed achieved was approximately 0.81 m/s. The flying control experiments showed diagonal legs ejecting alternately, with the robot completely leaving the ground. The duty factor of the support phase was calculated to be 0.31.
Limitations and Future Work
- Model Error: The paper acknowledges that modeling Billy's aspherical feet as spherical in the simulation introduces model error, contributing to discrepancies between actual and desired speeds.
- Parameter Tuning: While control parameters are proposed, their selection is often based on experimental analysis or animal data. The paper suggests simplifying parameter adjustment through online learning for future work.
- Future Work: The authors plan to verify the proposed method on a larger quadruped robot platform and investigate methods for simplifying parameter adjustment using online learning.
Summary updated