STEP: Stochastic Traversability Evaluation and Planning for Risk-Aware Off-road Navigation (2103.02828v2)

Abstract: Although ground robotic autonomy has gained widespread usage in structured and controlled environments, autonomy in unknown and off-road terrain remains a difficult problem. Extreme, off-road, and unstructured environments such as undeveloped wilderness, caves, and rubble pose unique and challenging problems for autonomous navigation. To tackle these problems we propose an approach for assessing traversability and planning a safe, feasible, and fast trajectory in real-time. Our approach, which we name STEP (Stochastic Traversability Evaluation and Planning), relies on: 1) rapid uncertainty-aware mapping and traversability evaluation, 2) tail risk assessment using the Conditional Value-at-Risk (CVaR), and 3) efficient risk and constraint-aware kinodynamic motion planning using sequential quadratic programming-based (SQP) model predictive control (MPC). We analyze our method in simulation and validate its efficacy on wheeled and legged robotic platforms exploring extreme terrains including an abandoned subway and an underground lava tube.

• The paper introduces STEP, a method for risk-aware off-road robot navigation combining uncertainty-aware mapping, tail risk assessment using CVaR, and kinodynamic motion planning.
• STEP utilizes Conditional Value-at-Risk (CVaR) to focus on minimizing the risk of rare but high-impact navigation failures, allowing dynamic adjustment of risk tolerance.
• Validated in simulations and real-world deployments on wheeled and legged robots, STEP demonstrated efficacy in balancing path optimality with safety in challenging environments.

Stochastic Traversability Evaluation and Planning for Risk-Aware Off-road Navigation

The paper introduces STEP (Stochastic Traversability Evaluation and Planning), a method designed to address the challenges of autonomous navigation in unstructured, off-road environments. Current robotic systems tend to perform well in controlled settings, but transferring these capabilities to uncertain outdoor terrains presents increased complexity due to elements like rough terrain, localization errors, and sensor noise. By incorporating risk assessment and kinodynamic motion planning, STEP aims to enhance the autonomy and safety of ground robots operating in such environments.

Approach Overview

STEP combines multiple innovative components:

1. Uncertainty-Aware Mapping and Traversability Evaluation: This component rapidly creates and analyzes uncertain 2.5D terrain maps to assess traversability. The maps aggregate sensor data to account for sensor noise, occlusion, and localization errors, providing a comprehensive view of navigational risks like slopes, rough terrain, and narrow passages.
2. Tail Risk Assessment Using Conditional Value-at-Risk (CVaR): Instead of minimizing only the expected traversability cost, STEP reduces the risks of rare but high-impact failures. By using CVaR, the approach focuses on the tail of the risk distribution, allowing for dynamic adjustment of risk tolerance levels based on external factors such as mission requirements.
3. Efficient Risk and Constraint-Aware Kinodynamic Motion Planning: Implemented through sequential quadratic programming-based (SQP) model predictive control (MPC), STEP formulates a nonlinear optimization problem that simultaneously minimizes traversability risks while respecting the robot's dynamic and kinematic constraints.

Key Contributions

The authors assert several contributions through the development of STEP:

• 2.5D Traversability Analysis: Incorporating localization errors and sensor uncertainties, STEP provides a more robust traversability evaluation that integrates diverse sources of risk.
• CVaR-Based Planning Framework: A unified framework that merges traversal risks with CVaR, facilitating flexible adaptation of risk levels during planning.
• Scalable and Real-Time MPC Architecture: By efficiently solving non-linear, risk-constrained problems, STEP enables high-speed, on-the-fly planning and adaptation in unpredictable terrains.
• Empirical Validation: The system's performance has been evaluated using wheeled and legged robots in high-risk environments such as underground tunnels and rugged lava tubes.

Numerical Evaluation and Real-World Deployment

Simulations demonstrated the system's capacity to balance path optimality with risk management, showing that varying the CVaR risk level ($\alpha$) allows a trade-off between path length and safety. Higher $\alpha$ values resulted in safer but longer paths.

In practical experiments, STEP was deployed on robots navigating challenging environments such as a cluttered abandoned subway and a naturally rough cave. These real-world trials confirmed the method's efficacy in evaluating traversability and dynamically adjusting risk levels based on CVaR metrics, allowing for adaptable and resilient navigation strategies.

Future Implications

The implications of STEP for robotics and AI extend to improving autonomy in hazardous environments, like extraterrestrial exploration or disaster response scenarios. The methodological framework integrates risk management directly into motion planning, providing a basis for future work to further refine risk metrics and tailor the approach for specific applications.

STEP’s introduction of CVaR into robotics could motivate further research into risk-sensitive AI systems capable of operating under uncertainty. Future work may investigate further improvements in computational efficiency that allow even faster adaptation or exploration of new terrain representations that accommodate more complex environmental features.

In conclusion, STEP presents a substantial advancement in risk-aware navigation, enabling more autonomous operation in environments previously deemed too unpredictable or dangerous for traditional robotic systems.