Pacific · Astrognosy AI

Signals Talk.
Beacon Listens.

Behavioral anomaly detection across signal streams — industrial machinery, robotic sensor fusion, and edge telemetry. Beacon reads vibration, IMU, encoders, cameras, and rack-floor sensors. Without rules, without retraining, without GPU compute.

F1 0.832

CWRU Bearings

<2ms

Latency

CPU

Edge Ready

Training Samples

Two Surfaces · One Engine

🏭 Industrial Signal Detection

Bearing faults, motor degradation, and process anomalies from vibration, RPM, temperature, and motor-current streams. Calibrate on healthy windows, then run live. No fault samples required.

BEARING_FAULTVIBRATION_RMSSPECTRAL_DRIFTCWRU F1 0.832

Open the live industrial demo →

🤖 Robotic Sensor Fusion

LIDAR, IMU, encoders, cameras, SLAM, and thermal streams fused into a single behavioral baseline. Obstacle, drift, and sensor-degradation anomalies surface in real time, on the edge CPU.

LIDAR_OCCLUDEIMU_JERKPOSE_DRIFTSENSOR_LAG

See the robotics monitor below ↓

Supported Signal & Sensor Inputs

📊

Vibration

RMS, kurtosis, spectral content

⚡

Motor Current

Load signatures and stator faults

📡

LIDAR

Point cloud density & occlusion events

🔄

IMU

Acceleration, jerk, and orientation drift

⚙️

Encoders

Wheel slip and drive anomalies

📷

Camera

Focus, exposure, and coverage gaps

🗺️

SLAM

Pose confidence and map consistency

🌡️

Thermal

Motor temp and actuator overload

What Beacon Detects

Environmental Hazards

LIDAR occlusion events, unexpected obstacle proximity, and point cloud density drops that indicate physical hazards the robot is navigating toward.

LIDAR_OCCLUDEOBS_CLOSEDENSITY_DROPSCAN_FAIL

Motion Anomalies

IMU jerk events, wheel slip from encoders, and actuation failures that indicate traction loss, payload shift, or mechanical degradation during operation.

IMU_JERKENC_SLIPPOSE_DRIFTACT_FAULT

Sensor Degradation

Camera blur, lens contamination, IMU drift over time, and LIDAR return rate drops — sensor health anomalies that compromise navigation quality before errors surface in downstream systems.

CAM_BLURIMU_DRIFTLIDAR_LOSSSENSOR_LAG

How It Works

Step 01

Fuse & Tokenize

Readings across all sensor streams are fused and mapped to a shared behavioral vocabulary. Each sensor contributes a token class — heterogeneous inputs become a unified behavioral sequence.

LIDAR+IMU → tokens

Step 02

Sliding Windows

Token sequences roll across time windows sized for the robot's operational cadence — from 10ms for fast manipulators to 1s for AMR navigation cycles.

W = 20 cycles

Step 03

Compare to Baseline

The current window's behavioral fingerprint is compared against the calibrated baseline. The structural shape of signal co-occurrence tells the story of whether the system is behaving normally.

structural divergence score

Step 04

Alert & Act

Divergence above threshold triggers an alert — classified by sensor stream of origin. The robot's safety system receives a structured signal, not just a raw sensor value.

divergence above threshold → HALT/WARN

Live Robotics Simulation

Sensor-fusion monitor for a single robot. For the industrial bearing fault demo with CWRU benchmark results, open the live industrial demo →

Sensor Fusion Monitor

Streaming

Sensor Log

Signals Talk.Beacon Listens.

Signals Talk.
Beacon Listens.