Research Paper

Core Innovation

This paper introduces JustDepth, a single-stage radar-camera depth estimator trained solely with radar, camera, and single-scan LiDAR data. It aggregates radar returns into a fixed-width 1D representation to decouple runtime from point count, uses a Height Fusion Block for modality fusion, and employs a lightweight GNN for global depth propagation, achieving significant speed and artifact reduction improvements.

Why It Matters

Accurate and fast depth perception is critical for autonomous systems to navigate safely and efficiently. Existing methods often suffer from high latency or require complex multi-stage processing, limiting real-world deployment. JustDepth's approach lowers inference time drastically while maintaining accuracy, enabling scalable and reliable perception in automotive and robotics applications.

Potential Customers & Pain Points

• Autonomous vehicle manufacturers – Need low-latency accurate depth perception
• Robotics companies – Require efficient sensor fusion for navigation
• ADAS providers – Seek scalable real-time environment sensing
• Smart city infrastructure – Demand reliable depth data for traffic monitoring.

Business Model

Licensing the JustDepth platform to automotive OEMs, robotics companies, and ADAS providers; offering integration support and custom development services.

Competitive Landscape

• Waymo
• Tesla
• Mobileye
• Luminar
• Innoviz

Implementation Challenges

• Integration complexity with existing vehicle sensor suites
• Robustness under diverse environmental conditions
• Competition from established multi-sensor fusion solutions

Validation Strategy

• Benchmark performance on public datasets like nuScenes
• Pilot deployments with automotive partners for real-world testing
• Iterative improvements based on field feedback and edge case analysis