Algorithms, Analysis Skills, Artificial Intelligence (AI), Autonomous Driving Systems, Data Analysis, Deep Learning, Improvement Metrics, Machine Learning, Ranking Technology, Reinforcement Learning, Relocation Services, Scalable System Development, Simulation, Vehicle Fleets
Founded in 2016 in Silicon Valley, Pony.ai has quickly become a global leader in autonomous mobility and is a pioneer in extending autonomous mobility technologies and services at a rapidly expanding footprint of sites around the world. Operating Robotaxi, Robotruck and Personally Owned Vehicles (POV) business units, Pony.ai is an industry leader in the commercialization of autonomous driving and is committed to developing the safest autonomous driving capabilities on a global scale. Pony.ai's leading position has been recognized, with CNBC ranking Pony.ai #10 on its CNBC Disruptor list of the 50 most innovative and disruptive tech companies of 2022. In June 2023, Pony.ai was recognized on the XPRIZE and Bessemer Venture Partners inaugural "XB100" 2023 list of the world's top 100 private deep tech companies, ranking #12 globally. As of August 2023, Pony.ai has accumulated nearly 21 million miles of autonomous driving globally. Pony.ai went public at NASDAQ in November 2024.
Responsibility
- Build scalable systems for training and fine-tuning large generative models that produce realistic, informative driving behaviors for evaluation and scenario coverage.
- Implement and iterate on RL-style methods: algorithms, reward / preference objectives, and training setups suited to high-fidelity, insightful behaviors in simulation-aligned workflows (closed-loop evaluation mindset).
- Ship deep learning solutions (including LLM / VLM where appropriate) that improve human-led triaging, automate high-volume workflows, and support nuanced analysis of self-driving behavior to surface critical anomalies.
- Own production-oriented ML for fleet-scale assessment: training, optimization, monitoring, and iteration of models used to judge performance across large real-world exposure.
- Design and evolve data + evaluation systems inspired by RL from human preferences (RLHF) and related paradigms-turning preference/judgment signals into repeatable, scalable training and evaluation loops.
- Partner broadly with teams such as Prediction, Planning, Research, and platform/engineering leads to land cross-cutting improvements with clear metrics.