BlackCat Robotics

TechMedix

Overview
AR Mode
Maintenance
Repair Intel...
Knowledge BaseCertifications
Fleet & Nodes
Robots & NodesDrones
Drone Fleet
Habitat
Data Centers
Network
Operations
DispatchOperations
Energy & Grid
Billing
Settings

Repair Intelligence

Knowledge Hub

The bridge between human technicians, robot hardware, and AI systems. Study failure modes, understand component anatomy, explore simulation environments, and take your certification exam — all from one place.

Platforms

20

Failure Signatures

73

Sim Environments

6

AI Layers

4

Layer 1 — Physical

Robot Platform Catalog

Each platform entry includes known failure signatures, severity, and maintenance context. Expand any platform to study its repair profile before your exam.

Humanoid

HumanoidFeatured Partner

Unitree G1

Unitree Robotics

43-DOF bimanual humanoid with Dex1 hands. VLA-capable via UnifoLM-VLA-0. Official TechMedix partner platform.

Total DOF

43

Arm DOF

7 per arm

F/T Sensors

13

Control Freq

20 Hz

Known Failure Signatures

●

Actuator Overheat

Joint temp > 75°C sustained > 30s

○

Joint Backlash

EEF tracking error > 15mm

○

F/T Sensor Drift

Static FT reading > ±2N / ±0.5Nm

5 signatures

Manual DiagramCertify
Humanoid

Unitree H1-2

Unitree Robotics

Full-size humanoid at 180cm. 40-DOF, extended battery for long-shift warehouse and logistics deployment.

Total DOF

40

Height

180 cm

Payload

30 kg

Battery

~4h operation

Known Failure Signatures

●

Actuator Overheat

Leg actuator temp > 80°C during sustained gait

○

Joint Backlash

Leg joint position error > 20mm

●

Battery Critical

SOC < 15% during active task

3 signatures

Manual DiagramCertify
Humanoid

Figure 02

Figure AI

Commercial humanoid for automotive assembly. 16 DOF hands, OpenAI-trained reasoning. BMW Group deployment.

Height

168 cm

Weight

70 kg

Battery

~5h

Hand DOF

16 per hand

Known Failure Signatures

○

Hand Gear Wear

Finger position error > 2mm on precision grip

●

Head Camera Drop

Primary vision system offline > 100ms

2 signatures

Manual DiagramCertify
Humanoid

Tesla Optimus Gen 3

Tesla

Tesla's 3rd-gen humanoid. 22 DOF hands, FSD-derived perception stack, designed for Tesla factory operations.

DOF

35+

Weight

57 kg

Battery

~8h

Hand DOF

22 per hand

Known Failure Signatures

○

FSD Stack Latency

Perception inference > 50ms

●

Battery Critical

SOC < 20% during active sequence

2 signatures

Manual DiagramCertify
Humanoid

Digit V5

Agility Robotics

Warehouse-specialized bipedal. Deployed at Amazon. Optimized for tote handling in dynamic logistics environments.

Payload

16 kg

Height

175 cm

Battery

~16h (hot-swap)

Speed

1.5 m/s

Known Failure Signatures

○

Ankle Backlash

Ankle dorsiflexion error > 1.5°

●

Knee Actuator Overheat

Knee joint > 76°C during incline traversal

2 signatures

Manual DiagramCertify
Humanoid

Phantom Mk1

Physical Intelligence

π0-powered general-purpose humanoid. Diffusion policy control. Research-to-production platform.

Policy

π0 Diffusion

DOF

38

Battery

~3h

Known Failure Signatures

○

Diffusion Stall

Policy inference > 100ms per step

○

Gripper Drift

End-effector grip force variance > 15N

2 signatures

Manual DiagramCertify

Industrial

Industrial

Unitree B2

Unitree Robotics

Industrial quadruped. 23 kg payload, IP67, outdoor-capable. Common in construction survey and infrastructure inspection.

Payload

23 kg

Protection

IP67

Battery

~4h

Speed

6 m/s

Known Failure Signatures

●

Leg Actuator Overheat

Leg joint > 72°C on rough terrain

○

Foot F/T Drift

Contact detection false positive rate > 5%

2 signatures

Manual DiagramCertify
Industrial

Boston Dynamics Spot

Boston Dynamics

Industry-standard inspection quadruped. Arm-optional, API-first, payload-extensible. Deployed in 50+ industries.

Payload

14 kg

Battery

~90 min

Protection

IP54

Speed

1.6 m/s

Known Failure Signatures

○

Hip Actuator Wear

Hip position error > 0.8° during gait

○

Foot Sensor Drift

Ground contact confidence < 85%

2 signatures

Manual DiagramCertify
Industrial

Amazon Proteus AMR

Amazon Robotics

Autonomous mobile robot for warehouse pod movement. LiDAR-only navigation, no camera dependence for safety.

Payload

750 kg (pod)

Speed

1.1 m/s

Battery

~8h (hot-swap)

Nav

LiDAR SLAM

Known Failure Signatures

●

Drive Motor Overheat

Drive motor > 70°C under heavy pod

●

LiDAR Dropout

LiDAR scan rate < 5 Hz

2 signatures

Manual DiagramCertify
IndustrialWatch List

reBot-DevArm

Seeed Studio

True open-source 6-DOF robot arm — hardware blueprints, full BOM, and 3D print files published on GitHub. Python SDK, ROS1, ROS2, Isaac Sim, and LeRobot compatible. Same simulation stack as ROBO-1. Pre-integration: monitoring adoption curve for full TechMedix onboarding.

DOF

6

SDK

Python, ROS1, ROS2, Isaac Sim, LeRobot

Source

Open — hardware blueprints, full BOM, 3D print files

GitHub

Seeed-Projects/reBot-DevArm

Known Failure Signatures

○

Joint Backlash

EEF position error > 2mm on repetitive trajectory

○

Servo Overheat

Joint servo > 70°C during sustained high-torque moves

2 signatures

Certify
IndustrialAI Platform

NVIDIA Jetson AGX Thor

NVIDIA

Flagship edge AI compute module for humanoid robots, surgical systems, and industrial AI. Blackwell GPU, 128GB LPDDR5X, 40-130W. Ships with JetPack 7 / Ubuntu 24.04 / CUDA 13.0. Developer kit $3,499, GA August 2025.

AI Compute

2,070 TFLOPS FP4

Memory

128GB LPDDR5X

GPU

Blackwell

Power

40-130W

Known Failure Signatures

●

JetPack/CUDA Driver Conflict

CUDA 13.0 driver mismatch after Orin-to-Thor migration breaks inference stack -- verify JetPack 7 compatibility before migration

○

NIM Microservice Cold Start

Cosmos Reason NIM container latency >5s on first inference call at edge -- pre-warm containers on boot

○

Thermal Throttle

Sustained load >100W triggers clock throttling -- verify thermal paste application and enclosure airflow

6 signatures

Certify
IndustrialAgtech

Aigen Element gen2

Aigen

100% solar-powered agricultural robot for chemical-free weed control at plant level. 350W panel, all-wheel drive, stereo depth vision (gen2), intelligent mesh fleet coordination. Featured by NVIDIA Robotics (April 2026). Deployed in California Central Valley.

Power

Solar + battery (350W)

Drive

All-wheel drive

Field Hrs

Up to 14h/day

Fleet Cov

200 acres/season

Known Failure Signatures

○

Weed Misclassification

Dense cotton canopy (late season) causes false negative rate >15% -- retrain vision model with seasonal field data

●

Mesh Network Dropout

Fleet communication loss in hilly terrain -- robot becomes unreachable, manual retrieval required

○

Solar Charging Shortfall

Multi-day overcast reduces daily operational range by 40-60% -- pre-charge backup batteries before extended cloudy periods

4 signatures

Certify

Drone

Drone

DJI Agras T50

DJI

50L agricultural spray drone. Dual atomized + broadcast spreading. Active phased-array radar obstacle avoidance.

Payload

50 L / 40 kg

Spread Width

9 m

Flight Time

~17 min (full load)

Radar

Active phased-array

Known Failure Signatures

●

Motor Overheat

Prop motor > 85°C (full payload)

●

Battery Critical

SOC < 15% mid-mission

2 signatures

Manual DiagramCertify
Drone

Skydio X10

Skydio

Enterprise inspection drone with NVIDIA Jetson-based autonomy and 4K/thermal dual sensor. Used in infrastructure inspection.

Camera

4K + thermal

Flight Time

35 min

Wind Resistance

15 m/s

Autonomy

NVIDIA Jetson

Known Failure Signatures

○

Thermal Sensor Drop

Thermal feed missing > 500ms

●

Autonomy Stack Stall

Obstacle avoidance latency > 200ms

2 signatures

Manual DiagramCertify
Drone

Zipline Platform 2

Zipline

Fixed-wing VTOL delivery drone. 160km range, autonomous dock-to-dock. Medical and commercial delivery.

Range

160 km

Payload

3.6 kg

Speed

128 km/h

Type

Fixed-wing VTOL

Known Failure Signatures

●

Tilt Motor Overheat

VTOL tilt motor > 90°C during transition

●

Range Battery Low

SOC < 25% beyond return threshold

2 signatures

Manual DiagramCertify

Delivery

Delivery

Serve RS2

Serve Robotics

Sidewalk delivery robot. LiDAR + camera fusion, 4-wheel drive. Deployed in LA and Dallas for food delivery.

Speed

1.5 m/s

Payload

18 kg

Battery

~12h

Sensors

LiDAR + 8× camera

Known Failure Signatures

●

Perimeter Camera Drop

Any of 8 cameras offline > 200ms

○

Wheel Encoder Drift

Odometry error > 5cm/m traveled

2 signatures

Manual DiagramCertify
Delivery

Starship Gen 3

Starship Technologies

Campus delivery robot. 6-wheel drive, 40+ cm obstacle clearance. Operates in 50+ cities.

Speed

1.8 m/s

Payload

10 kg

Battery

~6h

Cameras

12× wide-angle

Known Failure Signatures

●

Battery Critical

SOC < 12% away from dock

○

Drive Motor Overheat

Wheel motor > 65°C on incline

2 signatures

Manual DiagramCertify

Micromobility

Micromobility

Lime Gen 4 E-Bike

Lime

4th-gen shared e-bike. Swappable battery, IoT-connected, deployed in 200+ cities.

Motor

250W rear hub

Battery

36V swappable

Range

40+ miles

Connectivity

Proprietary cellular

Known Failure Signatures

○

Battery Cell Drift

±50mV cell delta — pack imbalance, swap recommended

●

Battery Thermal

>45°C at rest — thermal event risk

○

Hub Motor Bearing

Vibration >0.8g RMS — bearing wear onset

9 signatures

Manual DiagramCertify
Micromobility

Bird Three E-Scooter

Bird

3rd-gen shared scooter. Reinforced frame, 40km range, GPS geofencing. 350W motor with integrated non-swappable battery.

Motor

350W rear hub

Battery

36V 10.5Ah integrated

Range

25+ miles

Connectivity

Cellular + BLE

Known Failure Signatures

○

Battery Cell Drift

±50mV cell delta — pack imbalance

●

Battery Thermal

>60°C BMS cutoff — non-swappable pack at risk

○

Hub Motor Bearing

Vibration >0.8g RMS — bearing wear onset

10 signatures

Manual DiagramCertify
Micromobility

Rad Power RadCommercial

Rad Power Bikes

Commercial cargo e-bike. 150kg total load capacity, 750W motor, used for last-mile delivery fleets.

Motor

750W rear hub

Battery

48V 14Ah 672Wh

Range

45+ miles

Payload

120 lbs

Known Failure Signatures

○

Battery Cell Drift

±60mV cell delta — large 48V pack imbalance

●

Battery Thermal

>50°C at rest — 672Wh pack thermal event risk

○

Hub Motor Bearing

Vibration >0.8g RMS under cargo load

10 signatures

Manual DiagramCertify

Layer 1 — Physical

Component Anatomy

What is physically inside a robot — and how each part fails. Every entry includes a "human bridge" — how to feel, see, or measure the failure before you ever open a diagnostic tool.

Actuators

A typical humanoid uses 25-30 actuators split between rotary (shoulders, elbows, hips, knees) and linear (legs, torso). Each rotary actuator combines BLDC motor + harmonic reducer + encoder + torque sensor.

Key Fact

Tesla Optimus: 20 rotary + 14 linear actuators. Cost breakdown: reducer 36%, torque sensor 30%, motor 13.5%.

Human Bridge

Think of actuators as the robot's muscles -- they create movement at every joint. When an actuator fails, the robot loses range of motion in that limb.

Harmonic Drive Reducers

Supply Bottleneck

Strain wave reducers provide high gear ratios with near-zero backlash. Largest single cost driver (~36%) in rotary actuators. Only 12% of global machine tool manufacturers meet the precision requirements.

Key Fact

Harmonic Drive holds 20-25% global market share. Alternatives: cycloidal-pin gear, planetary gearbox.

Human Bridge

Reducers are why robots move smoothly rather than jerking. Backlash (slop in the gears) is a primary failure symptom -- you'll feel it as imprecision or vibration at the end-effector.

BLDC Motors

Brushless DC motors are the dominant choice across humanoids for high torque density in compact form factors. Controlled by ESC/FOC drivers. No brushes = longer service life vs brushed motors.

Key Fact

Dominant suppliers: Maxon (25-30% share), Kollmorgen (15-20%). China alternative: PMSM low-inertia high-speed motors (Unitree).

Human Bridge

If a joint runs hot or draws unusual current, the motor winding or controller (ESC) is your first diagnostic target. Oscilloscope the drive signal before replacing the motor.

On-Robot Compute

Largely standardized on NVIDIA Jetson (Orin, AGX Thor). Tesla is the exception with its proprietary AI5 SoC. Key metrics: TOPS, TOPS/watt, and memory bandwidth.

Key Fact

NVIDIA Jetson AGX Thor: 2,070 FP4 TFLOPS (Blackwell). Tesla AI5: proprietary. Chinese alternative: Horizon Robotics.

Human Bridge

The compute module is the robot's brain. Heat + throttling = inference slowdown = sluggish reactions. Check thermal paste and airflow before suspecting software issues.

Sensors

2-7 cameras per robot for perception. IMUs for orientation. Force/torque sensors on joints. LiDAR for mapping (Agility, Unitree, AGIBot). Tactile sensors on hands for dexterous manipulation.

Key Fact

10 of 13 major OEMs have tactile sensing. Sony and Intel RealSense dominate camera supply.

Human Bridge

A camera dropout causes immediate autonomy loss. IMU drift causes balance instability. F/T sensor miscalibration causes grip failures. Each sensor has a clear, testable failure signature.

Batteries

Lithium-ion and lithium-polymer packs. Operating times range 2-14 hrs. Cell imbalance (+/-50mV delta) is the primary aging indicator. XPeng is pushing toward all solid-state.

Key Fact

Capacity range: 0.84-5 kWh across humanoids. Charging modes: self-charge, hot-swap, wireless inductive.

Human Bridge

Battery health is the most measurable robot health signal. Monitor cell delta voltage, not just total SOC. A swollen pack is a thermal event risk -- never charge it.

End Effectors (Hands)

Range from 3-finger grippers to 22-DOF anthropomorphic hands. Drive types: tendon (1X, Tesla) vs motor+gear (Figure, Dexmate). Research standard: Shadow Robot 24 DOF, ~110K EUR.

Key Fact

Open-source option: ORCA Dexterity (17 DOF, $3.5K-$6.1K) from ETH Zurich spinoff.

Human Bridge

Hand repairs are the most tactile diagnostic task. Encoder drift, tendon tension loss, and fingertip sensor fouling each have distinct feel and response pattern. L2+ certification required.

Safety Standards

ISO 25785-1 (bipedal robots) is in working draft -- expected 2026-2027. ISO 10218 covers industrial robots. ISO 13482 covers service robots. EU Machinery Regulation applies Jan 2027.

Key Fact

Only Agility Robotics Digit has achieved NRTL field certification. AI Act applies to high-risk autonomous systems from August 2026.

Human Bridge

LOTO (Lock Out Tag Out) and zero-energy verification before any physical work. Know which standard covers the robot you're servicing -- it determines your liability and documentation requirements.

Layer 2 — Intelligence

Simulation Labs

Where humans get familiar with robot parts, failure scenarios, and toolsets before working on real hardware. These environments let you practice repairs, run fault injection, and understand how AI policies respond to component failures.

Industry Standard

NVIDIA Isaac Sim / Lab

Train humanoid locomotion and manipulation policies. Run thousands of parallel simulations. Integrates with ROS 2 and real hardware via Isaac ROS.

Engine

PhysX + Newton + MuJoCo

License

Free for R&D

Language

Python, C++

Cert Level

L3–L5

Open Repo / DocsLaunch Isaac Sim Docs
Research Default

MuJoCo

Fast contact physics. Standard benchmark environment for locomotion RL. Google DeepMind maintains it. MJX runs on GPU via JAX for massively parallel training.

Engine

MuJoCo (JAX backend via MJX)

License

Apache 2.0

Language

C, Python, JAX

Cert Level

L2–L4

Open Repo / DocsOpen MuJoCo Playground
Fastest Physics

Genesis

26 seconds to train a locomotion policy on an RTX 4090. Supports rigid, soft, fluid, and cloth simulation. Best for rapid policy iteration and diverse material interaction.

Engine

Custom multi-physics (rigid, MPM, SPH, FEM, PBD)

License

Apache 2.0

Language

Python

Cert Level

L3–L5

Open Repo / DocsGenesis Quick Start
NVIDIA / Berkeley / CMU

CaP-X (capgym/cap-x)

Benchmark + training framework where LLM/VLM agents write Python control code to operate physical robots. 187 tasks across Robosuite, LIBERO-PRO, and BEHAVIOR. CaP-RL takes a 7B model from 20% to 72% success in 50 iterations. 84% sim-to-real transfer on Franka Panda. Maps to TechMedix cert levels: L1=S1 tasks, L2=S2-S3+M1-M2, L3=S4+M3-M4.

Engine

MuJoCo + Isaac Sim (BEHAVIOR tasks)

License

Open Source

Language

Python

Cert Level

L1–L3

Open Repo / DocsBrowse CaP-X Tasks
Evaluation

Velxio

Under evaluation for TechMedix fault injection simulation track. Target use: technicians diagnose simulated hardware failures in a virtual environment before working on real robots.

Engine

Python / ROS 2 (evaluation)

License

MIT (evaluation)

Language

Python

Cert Level

L1–L2

Open Repo / DocsView Velxio Repo
Emerging Standard

Newton Physics Engine

NVIDIA + Google DeepMind + Disney Research convergence project. Bridges Isaac Sim and MuJoCo ecosystems. Expected to become the shared physics layer for robot training.

Engine

Newton (on NVIDIA Warp)

License

Open Source (Linux Foundation)

Language

Python, C++

Cert Level

L4–L5

Open Repo / DocsNewton on GitHub

Layer 2 — Intelligence

AI Intelligence Layer

How robots think, learn, and generalize — and what it means for diagnostics. Understanding the AI stack helps technicians know whether a failure is hardware, firmware, or policy-level.

World Models

Learn to predict how the physical world evolves in response to robot actions. Let robots practice tasks "in imagination" before acting in reality — dramatically reducing real-world data needs.

Examples

NVIDIA Cosmos, Wayve GAIA, Dreamer v3

Tech-to-Field Bridge

When a robot hesitates before a new task, it may be running a world model rollout to predict outcomes. Failure signatures: policy stalls, repetitive motions, unexpected stops.

VLA Models

Vision-Language-Action models unify perception, language grounding, and control policies. They are the generalist reasoning layer — the robot's 'common sense' for embodied tasks.

Examples

π0 (Physical Intelligence), OpenVLA, RoboFlamingo, Helix (Figure), UnifoLM-VLA (Unitree)

Tech-to-Field Bridge

VLA inference stall = motor overheat or perception dropout feeding corrupted state. Check camera feeds and motor temps before blaming the model.

Reward Models

Score video trajectories against language instructions. Provide the dense reward signal that VLA policies need for RL training, data filtering, and quality estimation.

Examples

Value-Order Correlation (VOC) scorers, trajectory preference models

Tech-to-Field Bridge

During policy training, reward model quality directly determines task success rate. A miscalibrated reward model produces robots that "game" the score without completing the task.

Simulation → Reality

The sim-to-real gap is the central challenge. Domain randomization (varying friction, lighting, mass) in simulation trains policies that transfer to real hardware.

Examples

Isaac Lab DR, MuJoCo domain randomization, Genesis material variation

Tech-to-Field Bridge

Policies that work in sim but fail on hardware usually have a sensor calibration mismatch or a joint friction model error. The fix is hardware measurement, not re-training.

Layer 3 — Human

Ready to certify your knowledge?

Five certification levels from Operator to Autonomous Systems Architect. Study the guide above, then take the exam directly — no GitHub required.

L1 OperatorL2 TechnicianL3 SpecialistL4 Systems Eng.L5 Architect
View CertificationsStart L1 Exam Now