
🤖 Robotics · Data Flow
How data from real robots and simulation is collected, labelled and used to train and deploy new robot models.
Drawing diagram…
Robot learning data flow: robots upload camera, lidar and log data from missions, especially failures. Data is stored in a data lake, selected and labelled. A simulator generates synthetic scenarios. The training pipeline trains perception and grasping models, an evaluation step tests them in simulation and on a test robot, and approved models are deployed to the fleet.
flowchart LR
subgraph EXT["External Entities"]
ROB[Robot Fleet]
LAB[Labelling Team]
end
subgraph PROC["Processes"]
P1[1. Upload mission data]
P2[2. Select interesting episodes]
P3[3. Label data]
P4[4. Generate simulation scenarios]
P5[5. Train models]
P6[6. Evaluate and approve]
end
subgraph STORE["Data Stores"]
D1[("Data Lake (S3)")]
D2[("Labelled Datasets")]
D3[("Model Registry")]
end
ROB -->|"1. Camera, lidar, logs"| P1 --> D1
D1 --> P2 --> P3
LAB -->|"2. Labels"| P3
P3 --> D2
P4 -->|"3. Synthetic data"| D2
D2 --> P5 --> D3
D3 --> P6
P6 -->|"4. Approved model"| ROBThe software that runs a fleet of warehouse robots: order intake from the WMS, task allocation, traffic control, charging and the operator dashboard.
The states of an autonomous mobile robot doing pick tasks: idle, navigating, picking, delivering, charging, and recovering from faults and blocked paths.
The software nodes inside a ROS 2 autonomous mobile robot: sensors, localisation, mapping, planning, control and the fleet connection.
How a robot arm picks items from a bin with vision guidance: detection, grasp planning, motion, placement and error handling.
The network inside a robotic work cell: robot controllers, PLC, safety systems, vision and the connection to the plant network.
A project plan for introducing autonomous robots into a warehouse: site survey, mapping, integration, pilot, training and scale-up.