Hardware & Physical AI Lab

D-CSIL's founder is an engineer with a systems-engineering background who prototypes hardware devices end to end: circuit and sensor selection, embedded firmware, data acquisition, and the software that makes the measurements mean something.

A two-unit BLE system, firmware v2.5 (March 2025), status: working great. An RV_Leveler sensor peripheral streams X/Y tilt and total error over a custom BLE service to an ESP32 T-Display handheld that renders a live bubble level on its TFT — graded status thresholds (LEVEL <0.5°, NEAR <1.5°, MID <3.0°, FAR), pitch-graded buzzer feedback (200/350/500 Hz), battery monitoring with charge detection, RSSI readout, a 60-second no-signal auto-shutdown, and deep-sleep wake on a hardware button. A companion Xcode app tracked the same peripheral live on Apple devices. Display-unit source is archived in the lab registry (docs/hardware/).

A monitor that measures the lab computer's vibration signature and operating temperatures to verify it is running within normal envelopes — instrumentation for the machines that train the models. Supporting artifacts on the bench: SensorTile firmware images, the SmartEdge Agile BLE telemetry dashboard (device firmware + real-time monitoring UI), and the always-on Pi fleet (Bailey serving Cortex, MQTT hive nodes).

Physical AI is the lab's stated direction: systems that observe environments through sensors, collect their own data, recognize events, and act — rather than only processing text that already exists. Every embedded device built here is a step toward models with real-world perception, and the CANIS concept's capture rig would be built on exactly this stack.