1. Algorithm R&D:
Design and implement multimodal fusion sensing algorithms (such as target detection, segmentation, tracking, slam, etc.) based on camera, lidar/millimeter wave radar.Optimize the performance and real-time performance of deep learning models (CNN/Transformer/PointNet + +, etc.) on image and point cloud data.
2. Data processing:
Build large-scale image and point cloud datasets, design data enhancement, annotation specifications, and automated processing processes.Solve practical problems such as sensor calibration, timing synchronization, and noise filtering.
3. Model deployment:
Deploy algorithms to embedded platforms (e.g. NVIDIA Jetson, Horizon, etc.) to optimize computational efficiency (TensorRT/quantization/pruning).
4. Technical research:
Research cutting-edge papers (such as Bev perception, 4D radar, Occupancy Networks, etc.) to propose innovative solutions to long-tail problems.
5. Cross-team collaboration:
Collaborate with sensor, regulatory, and system teams to drive algorithm productization.
1. Master's degree or above, majoring in computer/automation/information/electronics/software/robotics;
II. Knowledge and experience:
1. More than 3 years of experience in perception algorithm research and development, solid computer vision or point cloud processing background.
2. Proficient in PyTorch/TensorFlow, familiar with OpenCV/PCL/ROS and other tool libraries.
3. Familiar with classical algorithms (such as CNN series of well-known visual AI models, YOLO, PointPillars, CenterPoint, etc.).
III. Engineering Capability:
1. Proficient in Python/C + +, with embedded deployment experience (such as CUDA, ONNX).
2. Familiar with multithreading, GPU acceleration, communication protocol (can/Lidar protocol).
IV. Additional requirements (plus points):
1. Experience in autonomous driving L2 + or above projects, or top conference papers (CVPR/ICRA, etc.).
2. Familiar with multi-sensor fusion (camera + radar + IMU) or temporal modeling (RNN/3D convolution).
3. Familiar with AutoML, distributed training frameworks (such as MMDetection3D, OpenPCDet).