R2S2R Embodied AI Solution
Start in the real world.
Train in simulation, then return to reality.
Use a Real-to-Sim-to-Real (R2S2R) data feedback loop to turn real-world scenes into trustworthy 3D environments for training and evaluation. Deploy validated tasks back to the physical site, then use operational data to continually refine the environments, models and tasks.
R2S2R / REAL-TO-SIM-TO-REAL
Real | Reality capture
Capture photos, video, drone imagery and point clouds to reconstruct trustworthy 3D environments.
Sim｜Training and Validation
Configure objects, routes, events and tasks in digital environments, and record training and evaluation results.
Real | Deployment and feedback
Connect physical robots and on-site systems so operational data informs each round of improvement.
View R2S2R Demo
View a demonstration of real campus inspection tasks from scenario configuration to operation verification.
Key points for implementation
Putting embodied AI into practice requires an R2S2R data feedback loop
A one-off reconstruction or offline training session cannot account for every real-world change. MeTop.ai uses a shared spatial coordinate system to connect environments, tasks, evaluation, deployment and operational feedback.
01
Real scenes are difficult to reproduce
Dangerous, complex and long-tail scenarios are difficult to collect and verify repeatedly.
02
The training environment is separated from the on-site space
Maps, objects, routes, events and tasks lack unified coordinates and semantic associations.
03
Operational results are difficult to feed back consistently
Field test results are not retained as data for the next round of training, evaluation and task optimization.
Core R2S2R capabilities
From the real world to simulation training and back
Start with trustworthy 3D environments built through reality capture, then connect environments, objects, tasks, evaluation, deployment and operational feedback.
Transform the real environment into a computable trustworthy space
rebuild
Real: Reality capture and reconstruction
Use photos, video, LiDAR and drone data to build trustworthy 3D environments that can be explored, support localization and can be updated over time.
Make objects and relationships in the scene understandable by machines
Semantics
Sim: spatial semantics and object management
Configure areas, pathways, equipment, obstacles, task points, and risk boundaries to provide environmental context for the robot.
Configure and verify critical tasks before entering the field
training
Sim: task configuration and training verification
Configure tasks around inspection routes, target objects, abnormal events and emergency procedures, and support review of training processes and results.
Record training and task performance with unified metrics
Evaluation
Sim: Testing, evaluation and reporting
Record task status, events, trajectories and results to form traceable testing, training and operation reports.
Let the on-site running results enter the next round of optimization
closed loop
Real: On-site deployment and data feedback
Connect robots, tasks and on-site events, then feed operational results back into spatial, training and evaluation datasets.
R2S2R implementation process
From reality capture to deployment and feedback
01
Real | Capture and spatial reconstruction
Collect real environment data and complete 3D reconstruction, coordinate alignment and scene release.
02
Sim｜Semantics, tasks and training
Define objects, areas, routes, events and task rules, and conduct training, testing and evaluation.
03
Real | Deployment, operation and feedback
Integrate robot hardware or algorithms, then feed task records, on-site events and operational results into the next round of improvement.
Delivery content
What is included in the R2S2R embodied AI spatial solution?
The specific delivery scope is confirmed based on the scene scale, robot type, mission objectives, data conditions and deployment method.
01
Real scene collection and data specification
02
Trustworthy 3D environments and a semantic layer
03
Object, route, task and event configuration
04
Test evaluation and operation report
05
Interface integration and deployment documentation
Typical application scenarios
Start with a real space and a task
First complete the scene reconstruction and task configuration, and then gradually connect the robot body, algorithm and on-site system.
Application scenarios
Campus and factory inspections
For night inspections, equipment inspections, safety incident identification and task closed loops.
Application scenarios
High-risk environment rehearsal
Validate routes, tasks and incident response procedures before entering chemical facilities, mines or confined spaces.
Application scenarios
Low-altitude and unmanned operations
Support mission verification of drones, robot dogs and mobile robots in complex spaces.
Related solutions
Explore related spatial intelligence applications for embodied AI.
Next step
Use R2S2R to connect real space, training verification and field operation
Tell us the robot type, target scenario and mission requirements, and we will provide real scene collection, digital environment construction, mission verification and on-site access solutions.