锦鲤求职岗位描述
This role sits at the intersection of ADAS system architecture, perception technology, sensor fusion, embedded software, vehicle integration, and production delivery. The successful candidate will be responsible for defining the technical direction of perception and system solutions, guiding architecture decisions from concept to production, and ensuring that ADAS features meet performance, safety, robustness, scalability, and cost targets.
The position requires both strong hands-on technical depth and the ability to lead cross-functional engineering activities across perception, planning, control, systems engineering, validation.
Responsibilities
ADAS Perception and System Architecture
Define and own the perception and system architecture for ADAS L2, L2+, and L3 features.
Develop end-to-end architecture concepts covering:
Sensor suite configuration
Camera, radar, lidar, ultrasonic, and other sensor utilization
Perception algorithm structure
Sensor fusion strategy
Environment model design
Functional allocation between perception, prediction, planning, control, HMI, and safety layers
Translate feature-level requirements into system, subsystem, software, and component-level requirements.
Lead trade-off studies involving performance, cost, compute, packaging, latency, redundancy, scalability, and production feasibility.
Define scalable architectures that can support multiple vehicle platforms, trims, markets, and regulatory requirements.
Perception Technology Leadership
Provide technical leadership for perception functions such as:
Object detection and tracking
Lane, road edge, and free-space perception
Traffic sign, traffic light, and road marking recognition
Vulnerable road user detection
Sensor fusion and multi-object tracking
Occupancy grid or environment modeling
Localization-related perception interfaces
Guide the team in algorithm selection, architecture decomposition, performance optimization, and robustness improvement.
Work closely with AI/ML, computer vision, radar, LiDar, and embedded software teams to ensure perception solutions are production-ready.
Define perception performance metrics, KPIs, validation criteria, and acceptance thresholds.
Drive continuous improvement based on simulation, proving ground testing, public road data, corner-case analysis, and customer usage data.
Silicon & Platform Mapping: Partner with silicon vendors (e.g., NVIDIA Orin/Thor, Qualcomm Snapdragon Ride) to map complex neural networks to heterogeneous computing engines (CPU, GPU, NPU, DLA).
Compute Budgeting & Optimization: Define strict system budgets for latency, memory bandwidth, thermal limits, and PCIe/Ethernet throughput. Drive model optimization strategies, including quantization (FP8, INT8), pruning, and hardware-accelerated operator design.
China Urban Complexity: Tailor the perception architecture to handle unique, high-frequency edge cases in Chinese urban centers (e.g., high-density vulnerable road users/VRUs, aggressive cut-ins, non-standard vehicles, complex multi-level intersections, and construction zones).
Global Platform Scalability: Ensure the core architecture is modular and highly scalable, allowing seamless adaptation to overseas markets (e.g., high-speed highway autopilot in North America and Europe, compliance with UNECE regulations).
Data Loop & Compliance: Architect the on-vehicle shadow mode, trigger-based data harvesting, and edge-case filtering pipelines. Ensure all architectures comply with regional data privacy and security regulations (e.g., China Data Security Law, GDPR).
Hardware-Software Co-Design & Optimization
Silicon & Platform Mapping: Partner with silicon vendors (e.g., NVIDIA Orin/Thor, Qualcomm Snapdragon Ride) to map complex neural networks to heterogeneous computing engines (CPU, GPU, NPU, DLA).
Compute Budgeting & Optimization: Define strict system budgets for latency, memory bandwidth, thermal limits, and PCIe/Ethernet throughput. Drive model optimization strategies, including quantization (FP8, INT8), pruning, and hardware-accelerated operator design.
Global & China Market Localization
China Urban Complexity: Tailor the perception architecture to handle unique, high-frequency edge cases in Chinese urban centers (e.g., high-density vulnerable road users/VRUs, aggressive cut-ins, non-standard vehicles, complex multi-level intersections, and construction zones).
Global Platform Scalability: Ensure the core architecture is modular and highly scalable, allowing seamless adaptation to overseas markets (e.g., high-speed highway autopilot in North America and Europe, compliance with UNECE regulations).
Data Loop & Compliance: Architect the on-vehicle shadow mode, trigger-based data harvesting, and edge-case filtering pipelines. Ensure all architectures comply with regional data privacy and security regulations (e.g., China Data Security Law, GDPR)
Advanced Study and Technology Roadmap
End-to-End (E2E) AD Transition: Define the architectural roadmap to transition from modular "Perception->Prediction->Planning" stacks to unified, differentiable End-to-End deep learning architectures (e.g., UniAD, World Models). Architect the system data flows, closed-loop latency budgets, and safety fallback/guardrail mechanisms required for L2+ E2E deployments.
Feature & Query-Level Fusion: Architect robust multi-sensor fusion strategies integrating high-resolution Cameras, 4D Imaging Radars, Lidars, and Ultrasonics. Drive the transition from late (object-level) fusion to early/mid-level fusion (e.g., cross-attention Transformer fusion, deep sensor-fusion networks).
Degradation & Redundancy Management: Define sensor degradation models (e.g., lens blockage, adverse weather, sensor misalignment) and architect dynamic, fail-safe sensor-fusion fallback strategies to ensure high system availability.
Evaluation of industry advanced sensors & perception results and provide technical insights/guidance for downstream feature development team
Qualifications
Education Qualification
Master’s or Ph.D. degree in Computer Science, Robotics, Electrical Engineering, Artificial Intelligence
No. of Years of Experience8+Professional Exposure
(Technical Skills)
With background in computer science, Computer Engineering, Electrical Engineering, Electrical Computer Engineering, or Software Engineering
Preferred previous experiencesProven hands-on experience in ADAS/AD system architecture, perception software development, or deep learning deployment.
Proven track record of shipping production-grade L2+ assisted driving systems from concept to Start of Production (SOP).
End-to-End AD Experience: Direct experience in developing, training, or deploying End-to-End autonomous driving models (e.g., imitation learning, reinforcement learning, world models) or unified planning-oriented perception networks.
Data Engine & Auto-Labeling: Experience designing cloud-based auto-labeling pipelines, NeRF (Neural Radiance Fields) / 3D Gaussian Splatting for scene reconstruction, and massive-scale closed-loop simulation environments.
Industry Influence: Active contributor to the open-source AD community or author of peer-reviewed publications in top-tier AI/Robotics conferences (CVPR, ICCV, ECCV, ICRA, IROS)
Functional Skills
Deep Learning & Computer Vision: Expert-level understanding of modern AI architectures, including Vision Transformers (ViTs), BEVFormer, CNNs, 3D Object Detection, and Occupancy Networks.
Spatio-Temporal Modeling: Deep understanding of temporal sequence modeling (Transformers, LSTMs, 3D Convolutions) and spatial coordinate transformations.
Sensor Fusion: Strong theoretical and practical foundation in multi-sensor fusion (Kalman Filters, Hungarian Algorithm, Transformer-based cross-attention fusion).
Programming & Toolchains: Proficiency in PyTorch/TensorFlow, C++, Python, and CUDA. Deep experience with model deployment toolchains (e.g., TensorRT, ONNX, TVM).
Functional safety, SOTIF, and system reliability
Behavioral Skills
Excellent ownership for assigned deliverables and scope end-2-end.
Excellent organization, problem solving, tactical thinking, and time management skills.
Excellent communication skills.
Excellent written and verbal communication skills.
Special Knowledge and Skills RequiredPerception system design and sensor fusion
Computer vision, deep learning, and AI-based perception
Any Others
Object tracking, lane detection, free-space detection, and environment modeling
Make balanced decisions considering performance, safety, cost, timing, manufacturability, and customer experience.
Communicate complex technical topics clearly to both technical and non-technical stakeholders.
Proactively identify risks and drive mitigation plans