TOPSATR Embodied Intelligence The Foundation for Industrial AI Capabilities

In this talk

1. 0:00

   Building industrial digital twins

   TopStar and Loon AI developed a data platform for industrial handling using a physics engine and digital twins. The system recreates facilities, equipment, and operating processes.

2. 0:21

   Scaling simulation training

   Scan-to-simulation and generated scenarios create varied training conditions, including rare edge cases. The approach combines simulation pretraining with fine-tuning on physical robots, plus automated annotations checked by people. The stated goal is to reduce on-site debugging from months to weeks.

3. 0:46

   Connecting data and learning

   The platform brings simulation, robot, and production data together, and routes failures back into the training process. TopStar reports a tenfold improvement in data training efficiency.

4. 0:59

   Evaluating and deploying robot models

   An integrated AI framework covers perception, planning, and control, with incremental learning and model updates while running. Models are assessed quantitatively on success rate, efficiency, and safety.

5. 1:16

   Reusing the full development loop

   The data-to-deployment process is intended to support repeatable updates across embodied robots. TopStar says the approach can be adapted across industries and replicated at scale.