Autonomous Inspection and Data Collection System for Public Transportation Facilities

The transportation agency faces difficulties in conducting routine inspections of transit station infrastructure due to limited available infrastructure adjustments, tight operational time windows, and the need to efficiently process large volumes of visual data. Manual inspections are time-consuming, and existing methods lack automation, consistency, and rapid analysis capabilities, hindering timely maintenance and cleanliness assessments.

A large urban public transportation agency operating metro, bus, and paratransit services in a major city, responsible for daily passenger flows exceeding several hundred thousand.

• Develop an autonomous robot system capable of performing scheduled inspections of transit station areas after hours, ensuring comprehensive coverage with up to 86% efficiency comparable to manual inspections.
• Design a structured data collection and visualization platform to enable remote review and feedback, leading to improved anomaly detection accuracy (targeting around 70%) without human intervention.
• Automate identification of infrastructure issues such as lighting failures, trash, stickers, graffiti, and other anomalies to streamline maintenance dispatch, thereby improving cleanliness and safety.
• Create a feedback loop to iteratively enhance object detection models through user annotations and AI analysis, reducing false positives and negatives over time.
• Leverage collected data for trend analysis to inform operational decisions, staffing, and preventive maintenance strategies.
• Explore extending the system with additional sensors for air quality monitoring, infrastructure health assessment, and security improvements.

• Autonomous navigation within station environments, respecting operational constraints and manual access controls.
• High-resolution imaging sensors (including CAM+IR) for capturing comprehensive visual data from multiple angles at 10-meter intervals.
• Onboard data processing unit (equivalent to Spot CORE) for executing custom logic and real-time decision-making.
• Structured data storage and labeling system to organize thousands of images for ease of review.
• An internal analytics dashboard to visualize inspection results, anomalies, and historical trends.
• AI-powered object detection models trained for identifying trash, graffiti, lighting issues, and other infrastructure anomalies, with confidence scoring.
• Feedback collection interface for operators to refine AI models continuously.
• Capability to interface with dispatch systems for maintenance, cleaning, or security team alerts based on detected issues.

• Dispatch and maintenance management systems for task assignment based on inspection data
• Real-time data transfer modules for uploading captured data to central servers
• Feedback and annotation tools for continuous model improvement
• Existing station infrastructure access controls for safe navigation and inspection

• System should operate reliably during designated inspection windows (e.g., 1AM-4AM).
• Coverage goal of at least 86% of relevant station areas per inspection cycle.
• Object detection accuracy of approximately 70% in initial deployment, with iterative improvements.
• Data processing latency should allow timely dispatch of maintenance activities.
• System security measures to prevent unauthorized access or data breaches.
• Scalability to add sensors and functionalities without sacrificing performance.

By deploying an autonomous inspection system, the transportation agency aims to achieve approximately 86% station area coverage, improving inspection consistency and efficiency. The system's AI-driven anomaly detection with an estimated 70% accuracy will enable faster identification and dispatch of maintenance or cleaning crews. Over time, data-driven insights will support operational planning, resource allocation, and preventive maintenance, leading to increased safety, cleanliness, and overall passenger satisfaction.