AI-Powered GPS Mapping and Perception Stack for Autonomous Electric Vehicles
JKCOM is an open-source framework that combines computer vision, radar sensing, and GPS to create accurate, real-time HD maps and robust localization for self-driving electric vehicles - with a focus on safety, efficiency, and real-world deployment.
- Multi-Sensor Fusion: Camera (semantic segmentation, object detection) + Radar (velocity & range in all weather) + GPS/IMU
- HD Map Generation: Automatic creation and updating of high-definition maps (lanes, signs, obstacles)
- Real-time Localization: Visual + radar-enhanced positioning, even in GPS-denied environments (urban canyons, tunnels)
- Perception Pipeline: Object detection, tracking, free-space segmentation, and dynamic obstacle prediction
- Electric Vehicle Optimized: Energy-efficient routing, battery-aware path planning, and regenerative braking integration
- Simulation Ready: Compatible with CARLA, NVIDIA Isaac Sim, and ROS 2
- End-to-end perception → mapping → localization pipeline
- Radar-camera fusion models for adverse weather (rain, fog, night)
- AI-driven map feature extraction (lanes, traffic signs, road boundaries)
- GNSS-free fallback using visual localization
- Lightweight models suitable for edge deployment on vehicle compute units
- Modular design - easily swap sensors or add new modalities (LiDAR support planned)
- Computer Vision: OpenCV, YOLO, Segment Anything, Detectron2 / custom transformers
- Radar Processing: Radar point cloud processing + Doppler velocity
- Sensor Fusion: Kalman filters, Deep fusion networks, ROS 2
- Mapping & Localization: ORB-SLAM / Visual SLAM + HD map formats (OpenDrive, Lanelet2)
- AI Framework: PyTorch / TensorFlow with ONNX export
- Simulation: CARLA + ROS 2
- Deployment: Docker, NVIDIA Jetson / edge hardware support
While projects like Autoware provide full autonomous stacks, JKCOM focuses specifically on mapping + localization with a strong emphasis on camera + radar fusion for cost-effective autonomous electric vehicles. It's designed to be lightweight, open, and extensible for researchers, startups, and EV manufacturers in emerging markets (e.g., enabling autonomous buses in Enugu, Nigeria).
Open Source • Self-hosted • Real-time • EV-focused
Built for the future of sustainable autonomous mobility.
License: Apache 2.0
