Abstract
Adverse weather conditions and occlusions in urban environments result in impaired perception. The uncertainties are handled in different modules of an automated vehicle, ranging from sensor level over situation prediction until motion planning. This paper focuses on motion planning given an uncertain environment model with occlusions. We present a method to remain collision free for the worst-case evolution of the given scene. We define criteria that measure the available margins to a collision while considering visibility and interactions, and consequently integrate conditions that apply these criteria into an optimization-based motion planner. We show the generality of our method by validating it in several distinct urban scenarios.
Introduction
Autonomous vehicle research often assumes perfect sensor data, but real-world urban environments are fraught with occlusions—buildings, parked vehicles, and sharp corners—that create significant blind spots. If a self-driving car assumes the road is clear simply because no obstacles are detected, it risks collisions with hidden objects. This research addresses the challenge of proactive motion planning under limited visibility and sensor uncertainty. The goal is to formalize "cautious" driving behavior, such as creeping into blind intersections, into an optimization-based framework that ensures a safe fallback option is always available.
Methodology
The proposed framework utilizes an optimization-based motion planner that minimizes a cost function for comfort and progress while adhering to strict safety constraints.
Environment Modeling and Uncertainty
The vehicle’s position and speed are modeled as Gaussian distributions to account for localization and perception noise. This allows the planner to maintain a safety margin that accounts for the standard deviation of sensor measurements.
Visibility-Aware Constraints
A core contribution is the "Virtual Object" concept. In scenarios with limited visibility, the planner treats the edge of the visible range as a potential location for a hidden obstacle. For intersections, the system assumes a hypothetical vehicle might be approaching from an occluded area at the speed limit.
Fallback Planning
The planner ensures that for every planned trajectory, a valid "fallback" maneuver (typically full braking) exists. This maneuver must remain collision-free even if the environment evolves in a worst-case manner during the system's processing delay. This proactive approach forces the vehicle to slow down in anticipation of potential hazards rather than reacting only after they are detected.
Interaction Logic
The system incorporates the Intelligent Driver Model (IDM) to predict the intentions of other road users. Even when the ego vehicle has the right of way, it monitors for "uncompliant behavior." If another vehicle fails to slow down as expected for a yield sign, the ego vehicle proactively prepares to brake.
Results
The approach was evaluated in a proprietary closed-loop simulation environment, testing various sensor ranges and intersection occlusions.
- Visibility-Speed Correlation: Simulations demonstrated a direct link between sensor range and approach speed. For instance, reducing the sensor range from 110m to 30m caused the vehicle to significantly lower its speed to maintain safety margins.
- Smoothness and Safety: By using a continuous optimization-based planner (solved via the Ceres library), the vehicle achieved smooth velocity profiles. The planner successfully maintained collision probabilities below a defined threshold by incorporating sensor uncertainty into distance constraints.
Discussion and Conclusion
The research successfully formalizes cautious driving behavior through mathematical constraints. By treating visibility limits as hard constraints, the vehicle naturally exhibits human-like "creeping" behavior at blind intersections.
However, the current model primarily focuses on longitudinal control and relies heavily on accurate offline maps. Future work could extend this to lateral maneuvers, such as evasive steering, and incorporate more diverse road users like pedestrians and cyclists who have different dynamic constraints.
Summary updated