Challenges and Prospects of Autonomous Driving Hardware Development – Insights from Pony.ai

The session, presented by Li Lintao from Pony.ai, outlines the company's journey in autonomous driving hardware development, beginning with the first vehicle in early 2017 using a 64‑line LiDAR, progressing to a 32‑channel LiDAR plus multiple cameras in 2018, and culminating in the release of the Pony Alpha and Alpha2 hardware platforms.

Pony.ai maintains hardware teams in Beijing, Silicon Valley, and Guangzhou, comprising electrical, structural, embedded, FPGA, and vehicle engineers who collaborate across locations through weekly technical exchanges and project‑based division of labor.

The Pony Alpha2 system, released in Q4 2019, integrates upgraded sensor configurations (e.g., Hesai 64‑line LiDAR), a self‑developed sensor cleaning system for adverse weather, DBW (drive‑by‑wire) line‑control technology for multiple vehicle models, custom wiring harnesses, higher integration to reduce cabling, and a spacious trunk for robotaxi luggage.

Key challenges identified include:

Cost: sensor and positioning modules can cost hundreds of thousands of RMB, prompting efforts to custom‑design hardware and reduce unnecessary components.

Power and thermal management: vehicle ECUs typically consume ~10 W, while autonomous driving compute units exceed 1 kW, leading Pony.ai to develop redundant power supplies, heterogeneous computing with FPGA acceleration, and extensive thermal simulation and testing.

Rapid iteration: the company achieved three hardware generations within roughly 1.5 years, emphasizing fast technology adoption, flexible OEM collaborations, and streamlined internal review processes.

Mass production: scaling to millions of vehicles requires cost‑effective designs, modular manufacturing (CNC, sheet‑metal, casting), automated camera calibration pipelines, and robust production and quality control workflows.

Complex road scenarios: handling intricate intersections and diverse weather conditions demands continuous sensor upgrades and algorithmic improvements.

Overall, Pony.ai's approach combines in‑house hardware R&D, cross‑functional team collaboration, and iterative engineering to address the evolving demands of autonomous vehicle deployment.