Tech Foundation of WeRide and Bot.Auto - Tony Han & Xiaodi Hou

Why It Matters

High‑fidelity, AI‑driven simulation is becoming the decisive lever for scaling autonomous‑vehicle fleets and satisfying regulators, directly influencing industry economics and safety standards.

Key Takeaways

• WeRide's Genesis simulator uses generative AI for high-fidelity scenarios.
• Simulation must verify code and discover rare edge cases to be trusted.
• Sensor simulation remains limited; visual-only models lack LiDAR realism.
• Behavioral simulation can bypass sensor fidelity by modeling physical driving limits.
• Both firms stress moving from data‑driven to computing‑driven paradigm.

Summary

The MIT Mobility Forum session brought together Tony Han of WeRide and Xiaodi Hou of Bot.Auto to dissect the technology backbone of today’s autonomous‑vehicle push. Their conversation centered on four pillars—simulation and world models, human‑in‑the‑loop versus full autonomy, hybrid AI‑deterministic architectures, and safety evaluation for regulators—while also touching on business trajectories.

Han unveiled WeRide’s Genesis simulation platform, a generative‑AI‑powered engine that merges high‑fidelity physics with human‑like decision agents. By recreating urban environments in minutes and compressing millions of real‑world kilometers into days of virtual testing, Genesis claims to surface rare edge cases, verify code changes, and support L2++ through L4 sensor stacks. Hou countered with a cautionary tale of a sudden LiDAR blackout in Texas fog, illustrating the current limits of sensor simulation, especially for non‑visual modalities, and arguing that visual‑only synthetic data offers diminishing returns.

Key moments included Han’s analogy that a good simulator should “find the theory of relativity” when data are stripped away, and Hou’s observation that behavioral simulation—modeling physical limits of driver actions—can sidestep the need for perfect sensor fidelity. Both emphasized a shift from purely data‑driven pipelines to a “computing‑driven” paradigm where synthetic feedback loops replace costly real‑world miles.

The discussion signals that scalable, trustworthy AV deployment will hinge on simulators that both validate software and generate novel scenarios, while regulators will increasingly rely on such virtual evidence. Companies that master this balance stand to accelerate commercialization, reduce testing costs, and shape future policy frameworks.

Original Description

The autonomous vehicle industry has reached its performance inflection in 2026. The MIT Mobility Forum #167 brings together two architects of the autonomy landscape:

•Dr. Tony Han: CEO and Founder, WeRide

•Dr. Xiaodi Hou: CEO and Founder, Bot.Auto

We will explore the technical foundation of WeRide and Bot.Auto:

1. The role of simulators and world model

WeRide: GENESIS unifying physical and generative AI

Bot.Auto: Foundation Model-to-All Architecture

High-fidelity modeling of interactions with road users

2. "Human-in-the-loop" vs. Strictly "humanless" operation

3. Hybrid architecture: end-to-end AI model + deterministic safety layer

4. Safety evaluation and decision support for legislation