Data Integration and Digital Twins: SOBO Lab

To bridge the gap between simulation and reality, we’ve developed SOBO Lab (Simulated-Observed Bridging Operations Lab), an internal toolkit for incorporating real-world spatial data into our virtual environments.

Key Capabilities:

• Point Cloud Data Fusion: By leveraging spatial data from advanced sensors such as depth cameras, SOBO Lab enables the generation of high-resolution, real-world-inspired digital twins of physical environments.

• Multimodal Sensor Simulation: SOBO Lab integrates LiDAR, RGBD cameras, and pressure sensors into the simulation pipeline, allowing robots to be trained on diverse sensor data.

• Sim2Real Transfer Optimization: By reducing the “reality gap,” SOBO Lab ensures that policies trained in RoboGym translate effectively to real-world tasks. This includes the use of domain adaptation methods and noise injection techniques to simulate sensor and actuation variability.

Example:

• In Phase 2, real-world depth scans were processed through SOBO Lab to generate dynamic, point cloud-based virtual environments. This approach allowed rovers to train on Martian-like terrains that mirrored real-world geological structures, with an environmental sensing accuracy improved by 45.2%.