About The Company

DiDi's autonomous driving unit was established in 2016 with the mission of developing Level 4 autonomous driving (AD) technology to make transportation safer and more efficient. In August 2019, the unit became an independent company, DiDi Autonomous Driving, dedicated to advanced AD R&D, product application, and business expansion. We believe integrating AD technology into a shared-mobility fleet will generate immense social value. By leveraging DiDi's specialized technology, operational expertise, and integrated ecosystem, we are positioned to build and operate a highly efficient, user-oriented autonomous fleet.

About The Role

Autonomous driving software today improves mostly by hand: engineers collect data, train models, and tune modules. This role explores a different approach, in which AI agents take part in that work. Agents propose changes to policy code, data, training recipes, or model code. They then evaluate the results, diagnose failures, and turn verified fixes into reusable skills and data for the next round. With DiDi Voyager's real-world driving data and large-scale evaluation infrastructure, the goal is a loop that produces improvements that generalize across scenarios, pass safety gates, and reach the vehicle. The role combines machine learning, agentic systems, and autonomous driving, and will help shape both the onboard architecture and how driving software is developed.

Responsibilities

• Build agentic harnesses for autonomous driving. Agents get permissioned access to the driving stack, data, training infrastructure, and evaluation environments, along with automatic scenario setup and pass/fail checks, so they can run experiments without a person in every step. 
• Develop recursive self-improvement (RSI) loops. These include failure mining, candidate generation, evolutionary search, and targeted data generation aimed at the policy's current weaknesses. Each round should also improve the process that produces the next one.
• Curate a reusable skill library. Verified fixes become knowledge that later tasks can build on, with checks that a skill actually helps before it is promoted.
• Train LLMs on our own agent data. Turn development trajectories into training data, and fine-tune LLMs to work fluently with our software stack, tools, and codebase.
• Advance agent-as-policy approaches. Agents improve driving behavior directly: they write policy logic, curate data, adjust training recipes and objectives, and trace driving failures back to their causes.
• Shape the onboard architecture. Decide what runs on the vehicle and what stays offline, within onboard limits on latency, compute, determinism, and safety.
• Work closely with the prediction, planning, simulation, and onboard teams to turn prototypes into production workflows.

Qualifications

• Strong background in machine learning, such as deep learning, reinforcement learning, or decision-making systems
• Hands-on experience building LLM agent systems, such as tool use, multi-agent orchestration, or automated evaluation loops
• Experience training or fine-tuning LLMs for agentic workflows, including trajectory data curation, supervised fine-tuning, or RL from execution feedback
• Strong Python skills and working C++
• Rigorous experimental habits: controlled comparisons, held-out evaluation, and a clear line between what was measured and what was assumed
• Ability to own a system from research prototype to production

Preferred Qualifications 

• Built an agentic harness or self-improving loop for a specific product or domain
• Experience in autonomous driving or robotics, especially planning, prediction, simulation, or sim-to-real transfer
• Familiarity with code-as-policy, program synthesis, LLM-driven optimization, or evolutionary search
• Publications at top ML or robotics venues (NeurIPS, ICML, ICLR, CVPR, RSS, CoRL, etc.)
• BS, MS, or PhD in Computer Science, Machine Learning, Robotics, or a related field

The annual base salary range for the Senior ML Engineer position is $169,783–$282,264. In addition to base salary, compensation includes bonus, equity, and benefits. The salary ranges are determined by role, level, and location. Individual compensation within the range depends on several factors, including work location, job-related skills, relevant experience, and education or training.

I acknowledge that prior to submitting this application, I have read and accepted the Privacy Notice for California Residents which is available on https://v.didi.cn/AQnxlBa