ROS 2 Inspection & Maintenance Robotics Guide 2026

Inspection is the highest-ROI commercial robotics use case today — robots checking pipes, power lines, turbines, and bridges that are dangerous or tedious for humans. This guide covers the ROS 2 stack for repeatable autonomous inspection and anomaly detection.

1. The Inspection Robot Stack

Inspection value comes from repeatability: the same route, the same viewpoints, every time:

sudo apt-get install -y   ros-humble-nav2-bringup   ros-humble-robot-localization   ros-humble-image-pipeline

# Layers:
#   1. Localization      (map-based, sub-meter repeatable)
#   2. Mission executor   (fixed waypoints + inspection actions)
#   3. Data capture       (RGB, thermal, acoustic, gas)
#   4. Anomaly analytics   (compare vs baseline, flag defects)
#   5. Reporting          (geolocated findings -> CMMS)

2. Repeatable Inspection Missions

Define named inspection points with the sensor action to perform at each:

INSPECTION_MISSION = [
    {'id': 'gauge_A1', 'pose': (12.4, 3.1, 1.57), 'action': 'read_gauge'},
    {'id': 'valve_B2', 'pose': (18.0, 5.2, 0.0),  'action': 'thermal'},
    {'id': 'pipe_C3',  'pose': (22.5, 5.2, 3.14), 'action': 'visual+acoustic'},
]

def run_mission(self, mission):
    results = []
    for pt in mission:
        self.nav.goToPose(to_pose(pt['pose']))
        self.nav.waitUntilNavComplete()
        self.settle()                      # damp motion for a sharp capture
        results.append(self.capture(pt['id'], pt['action']))
    return results   # identical viewpoints enable time-series comparison

3. Reading Analog Gauges

Many facilities still have analog gauges. Read them with vision:

import cv2, numpy as np

def read_gauge(image, center, min_angle, max_angle, min_val, max_val):
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    edges = cv2.Canny(gray, 50, 150)
    lines = cv2.HoughLinesP(edges, 1, np.pi/180, 80,
                            minLineLength=40, maxLineGap=5)
    needle = longest_line_through(lines, center)   # the pointer
    angle = line_angle(needle, center)
    frac = (angle - min_angle) / (max_angle - min_angle)
    return min_val + frac * (max_val - min_val)     # gauge reading

4. Thermal Anomaly Detection

Overheating is an early failure signal. Flag hotspots against a baseline:

def detect_hotspots(thermal_frame, baseline, threshold_c=15.0):
    delta = thermal_frame - baseline           # rise vs healthy baseline
    hotspots = np.argwhere(delta > threshold_c)
    findings = []
    for (v, u) in cluster(hotspots):
        findings.append({
            'pixel': (u, v),
            'temp_rise_c': float(delta[v, u]),
            'severity': 'critical' if delta[v, u] > 30 else 'warning'
        })
    return findings   # a bearing running 30C hot = schedule maintenance

5. Visual Defect Detection

Detect corrosion, cracks, and leaks with a trained model, then localize in 3D:

def detect_defects(self, rgb, depth, camera_info):
    dets = self.defect_model(rgb)              # corrosion/crack/leak classes
    findings = []
    for d in dets:
        p3d = pixel_to_3d(d.u, d.v, depth, camera_info)
        world = tf_transform(p3d, 'camera', 'map')
        findings.append({'type': d.label, 'conf': d.conf,
                         'location': world})    # geolocated defect
    return findings

6. Change Detection Over Time

The real power: compare today's capture to the last run at the exact same viewpoint:

def compare_runs(current, previous):
    # Same pose => aligned images; diff reveals new/growing defects
    aligned = register(current.image, previous.image)
    diff = structural_diff(aligned, previous.image)
    growth = [r for r in diff.regions if r.area > MIN_CHANGE]
    return {'new_defects': growth,
            'trend': 'worsening' if growth else 'stable'}
    # Corrosion that grew 20% since last month -> prioritize repair

7. Confined & Hazardous Spaces

• Gas detection: carry gas sensors; abort and alert on dangerous readings.
• Tethered comms: RF fails in pipes/tanks — use tethers or store-and-forward.
• Intrinsically safe: ATEX/IECEx-rated hardware in explosive atmospheres.
• Self-recovery: plan egress so the robot never strands where a human can't reach.

8. Docking & Autonomous Recharge

# Scheduled autonomous inspection with self-charging
def daily_cycle(self):
    while True:
        wait_until(next_scheduled_run())
        self.undock()
        results = self.run_mission(INSPECTION_MISSION)
        self.publish_report(results)
        self.return_to_dock()
        self.charge_until(0.95)
    # Unattended 24/7 inspection is the commercial payoff

9. Reporting & CMMS Integration

• Geolocated findings: every defect carries a map/world coordinate.
• Severity triage: critical hotspots page on-call; minor ones queue.
• Work orders: push findings into the CMMS (Maximo, SAP PM) automatically.
• Audit trail: timestamped imagery per asset for compliance.

10. Real Platforms & ROI

• Spot + payloads: substation, plant, and site inspection.
• ANYmal: offshore and industrial autonomous inspection.
• Crawlers: pipe and tank internal inspection.
• ROI: removes humans from confined-space/height risk and enables continuous monitoring vs periodic manual checks.

Key Takeaways

Inspection robotics turns repeatability into value: fixed missions, identical viewpoints, and time-series comparison that catches problems while they are cheap to fix. Combine gauge reading, thermal hotspot detection, and vision-based defect localization, then compare each run to the last to spot growth. Autonomous docking enables unattended 24/7 coverage, and CMMS integration turns findings into work orders — which is exactly why inspection is one of the strongest commercial cases in robotics today.