UBTech Launches Thinker-Sim: Real Scenes Into Simulation, Robots Train Virtually First

What happened

The official introduction names three industry pain points the platform addresses: high real-world data collection costs, complex simulation environment construction and long algorithm validation cycles. Thinker-Sim puts three capabilities in one system: digital replication of real scenes, automatic large-scale simulation data generation and standardized robot performance evaluation, positioned as a one-stop embodied-AI technology base.

The scene-to-simulation chain has two steps: the in-house Thinker-GS high-fidelity scene reconstruction algorithm and full-process SimReady asset auto-generation technology, linking real-scene capture, 3D reconstruction and simulation asset generation. The company says a 3D reconstruction of a real factory line can reproduce the plant's spatial layout, environment structure and visual detail, and load robots, production equipment and people as foreground assets per training, data collection or evaluation needs; commercial-service scenes can likewise get digital-twin environments with realistic visuals quickly.

Two designs lower the entry barrier: built-in standardized simulation task templates let users start a simulation task via template selection and parameter settings without building an environment from scratch, with real-time video streaming, task status monitoring and result feedback during simulation; and the in-house Ulab-Agent takes natural language as a single interaction entry, calling underlying atomic skills for scene creation, environment configuration, randomization, task orchestration and scheduled execution on its own. UBTech says flexible configuration of scene, robot, task, algorithm and evaluation modules fits more than 90% of target simulation scenarios, a company statement.

Key facts

1. Publisher and date: UBTech, Nov 2, 2026
2. Product: Thinker-Sim embodied-AI simulation platform, built on the Real2Sim2Real framework
3. Three capabilities: digital replication of real scenes, large-scale synthetic simulation data, robot capability evaluation
4. Scene reconstruction: in-house Thinker-GS high-fidelity reconstruction algorithm and full-process SimReady asset auto-generation
5. Task templates: built-in standardized templates, no environment built from scratch; real-time video streaming, task status monitoring and result feedback
6. Interaction entry: in-house Ulab-Agent takes natural language as a single entry and calls atomic skills for scene creation, configuration, randomization, orchestration and execution
7. Coverage claim: UBTech says flexible module configuration fits more than 90% of target simulation scenarios, a company statement