Launch HN: Salem Robotics (YC S26) – Software for industrial inspection robots

Hello Hacker News! We are thrilled to introduce Salem Robotics, a part of the YC S26 batch. We are building the intelligence layer that transforms standard mobile robots into high-precision industrial inspection assets.

The Problem: The "Human-in-the-Loop" Bottleneck

For decades, industrial inspection (checking for gas leaks, structural cracks, or thermal anomalies) has relied on humans walking through dangerous environments with handheld sensors. This approach is:

• Inefficient $\rightarrow$ Slow and labor-intensive
• Safe $\rightarrow$ Inherently risky for personnel
• Consistent $\rightarrow$ Prone to human error and oversight

"The hardware for robotics has finally caught up, but the software required to make a robot actually understand a complex industrial site is still missing. We are filling that gap." — Founder, Salem Robotics

---

Our Solution: The Salem OS

Salem Robotics provides a hardware-agnostic software suite that allows companies to deploy autonomous inspection fleets without writing a single line of low-level navigation code.

Key Features

1. Semantic Mapping: Our software doesn't just see a "wall"; it identifies a Pressure_Valve_v4 or a Steam_Pipe_Joint.
2. Anomaly Detection: Using edge-AI, we detect deviations from the "golden state" in real-time.
3. Seamless Integration: We plug directly into existing ERP and maintenance systems.

Comparison: Manual vs. Salem-Powered Inspection

---

How It Works: The Technical Pipeline

Our system operates on a closed-loop feedback mechanism. The robot captures raw data, the Salem engine processes it, and the output triggers a maintenance ticket if a threshold is exceeded.

Implementation Example

To trigger a specific inspection routine via our API, you can send a simple JSON payload to the robot's controller:

POST /api/v1/inspection/start

{
  "robot_id": "unit_04",
  "zone": "boiler_room_alpha",
  "sensors": ["thermal", "acoustic", "lidar"],
  "priority": "high",
  "callback_url": "https://webhook.site/maintenance-alert"
}

---

The Math Behind the Detection

We utilize a proprietary reliability score to minimize false positives. The probability of a true defect $P(D)$ given a sensor alert $A$ is calculated using a modified Bayesian approach:

$P(D|A) = \frac{P(A|D) \cdot P(D)}{P(A)}$

Where:

• $P(A|D)$ is the sensitivity of the sensor.
• $P(D)$ is the historical failure rate of the component.

---

Current Status & Roadmap

We are currently in private beta with three Fortune 500 energy companies. Here is what we are tackling next:

• Basic SLAM integration for indoor environments.
• Thermal anomaly detection module.
• Multi-robot coordination (Swarm Inspection).
• Integration with Digital Twin platforms.
• Support for outdoor, all-weather chassis.

Figure 1: The Salem Dashboard showing real-time heatmaps of a facility.

We would love to hear your thoughts on edge-compute vs. cloud-processing for high-bandwidth LIDAR data. Ask us anything!