Robotics

Du capteur au plan de vol — en boucle fermée.

Autonomy stack

• Localisation & mapping — SLAM, visual-inertial odometry, factor-graph fusion.
• Perception — LiDAR, radar, vision, multimodal sensor fusion; foundation-model perception.
• Prediction — motion forecasting, scene completion, intent modelling.
• Planning — classical motion planning, lattice planners, learned policies, hybrid stacks.
• Control — MPC, learned controllers, safety filters.

Runtime

• ROS 2 — managed nodes, DDS bridging, secure transport, lifecycle management.
• NVIDIA Isaac — sim-to-real pipelines, Isaac ROS integration, hardware-in-the-loop.
• Edge inference — ONNX, TensorRT, OpenVINO for on-device perception.
• Real-time — RT-PREEMPT Linux, dedicated cores, deterministic networking.

Simulation

• NVIDIA Omniverse / Isaac Sim — physically-accurate sim for autonomy development.
• Gazebo, Webots — open-stack sim for research and validation.
• CARLA, AirSim — driving and aerial autonomy sim.
• Domain randomisation — sim-to-real transfer with parametric scene generation.

Hardware-in-the-loop

Standing benches for HIL testing — flight controllers, vehicle ECUs, manipulator controllers, sensor stacks. Wired to the simulation backend so the same scenarios run sim-only, HIL, and live with identical orchestration code.

Fleet management

• OTA firmware and ML-model updates — signed, attested, rollback-capable.
• Cryptographic identity per platform — one cert, one key, in HSM or TPM.
• Telemetry pipelines — sensor streams to S3, summaries to ClickHouse for fleet analytics.
• Mission planning & replay — integrated with the planning stack.

Defence and mission robotics

For federal and allied-defence customers, robotics workloads run on Osage Cloud’s sovereign-deploy substrate — air-gapped, attested, FIPS-aligned, with cryptographic provenance on every sensor frame and every actuation command. Composes with the federal capability statement at osage.global/capabilities.

Robotics desk: [email protected].