TL;DR
This paper presents a novel safety-critical control framework for quadrupedal robots navigating complex, multi-tiered industrial environments like distillation columns. By integrating trajectory generation, safety-critical control using Control Barrier Functions (CBFs), and adaptive gait transitions, the system ensures robust and safe locomotion, even when encountering unexpected obstacles or unsafe regions.
What Problem Does This Paper Address?
Industrial inspection and maintenance tasks in environments like distillation columns pose significant challenges for robots. These environments are often unstructured, hazardous, and feature multiple levels or tiers, requiring robots to not only navigate complex terrains but also safely transition between these levels. Existing research on quadrupedal robots has largely focused on locomotion on a single level or general obstacle avoidance. However, there's a critical gap in developing comprehensive control frameworks that specifically address the safety-critical aspects of multi-layered navigation, including seamless transitions between levels and dynamic adaptation to unsafe areas within these confined spaces. This paper aims to fill this gap by proposing a unified approach for safe, autonomous operation in such challenging industrial settings.
Methods
The authors propose a multi-layered architectural framework that combines several key components:
- Trajectory Generation: For seamless transitions between column levels, they employ an optimization-based approach (using FROST) to generate joint trajectories. This considers the robot's full-body dynamics and constraints, ensuring smooth upward and downward movements through different contact phases (e.g., all feet, rear feet, front feet).
- Safety-Critical Control (CBFs): Control Barrier Functions (CBFs) are central to ensuring safety. They use an ellipsoidal form of CBF to define safe regions and dynamically adjust the robot's path (hpath) and gait (hgait). This allows the robot to avoid unsafe areas, such as the "manway" in the column, by modifying its control input.
- Footstep Re-planning: When a planned foothold falls within an unsafe region, a geometric re-planning strategy is employed. This strategy identifies the nearest safe point to the original foothold, ensuring the robot maintains stability and avoids collisions. This is applied to the full-body model.
- Gait Transition: The robot dynamically switches between different gaits (e.g., diagonal-gait trot and quasi-static walking) based on the safety level assessed by the CBF. If the robot approaches an unsafe region (hgait < 0), it transitions to a more stable, slower gait.
- Low-Level Controller: A comprehensive optimization-based QP controller is used to compute control commands for the robot's full-body dynamics, taking into account contact constraints, friction cones, and input limits.
Datasets
Not discussed in the paper. The experiments were conducted on a real quadrupedal robot platform.
Key Insights and Ideas
- Unified Safety Framework: The paper demonstrates that integrating trajectory generation, CBF-based safety filtering, and footstep re-planning within a layered architecture is crucial for safe multi-layered locomotion.
- Adaptive Gaits for Safety: Dynamically switching gaits based on safety assessments (via CBFs) is a powerful mechanism for ensuring stability in hazardous or confined spaces. A slower, more stable gait is preferred when approaching unsafe regions.
- Geometric Safety for Rectangular Obstacles: Using an ellipsoidal CBF to define safety around rectangular obstacles (like the manway) is an effective adaptation of CBF theory for practical industrial scenarios.
- Synergy of Reduced-Order and Full-Body Models: The framework effectively leverages a reduced-order model for safety-critical control (CBFs) and a full-body model for trajectory generation and low-level control, bridging the gap between simplified safety guarantees and realistic robot execution.
Key Results
The paper presents successful hardware experiments demonstrating:
- Stable Level Transitions: The robot successfully performed both downward and upward transitions between column levels, with transition durations of 8 seconds. Visualizations and joint position data confirmed smooth, collision-free movements.
- Safety-Critical Locomotion: The robot navigated a confined space around a manway, activating the CBF-based safety filter to avoid the unsafe region. It demonstrated dynamic gait transitions from a trot to a quasi-static walk when entering the unsafe zone.
- Effective Foothold Re-planning: The footstep planning was adjusted in real-time to steer clear of the manway, with modifications to left-sided foot positions clearly illustrated in experimental results.
Limitations and Future Work
- Model-Based Approach: The current framework relies on a model-based approach. The authors acknowledge that uncertainties stemming from model inaccuracies can be a limitation.
- Future Work:
- Perception and Localization: Augmenting the framework with perception and localization components to mitigate uncertainties from the model-based approach.
- Full Autonomy: Deploying the robotic system with full autonomy, governed by a state-machine, for comprehensive inspection missions in multi-tiered environments.
Summary updated