The Hardware Layer

Training starts with
what the robot is made of.

Trained behaviour is the product — but behaviour can only be trained on hardware that holds up. This is the electronic layer we design against and the supply reality we plan around: three module families, the semiconductor content of a single platform, and the three components most likely to constrain a roadmap.

Discuss a Platform See the Numbers
3Module families
60–160Power-stage units per platform
~22%Sensor share of BOM
<30%Force-sensor local penetration
Module families

Three families, one interface philosophy

The same families cover industrial arms, quadrupeds, AMRs and humanoids. What changes across those platforms is quantity and thermal duty — not the way the interface is defined.

NRB-Drive

Integrated joint drive control

MCU, current sensing, gate driver and power stage qualified together on a single board, with the thermal path taken to the housing.

  • Configurable joint count
  • EtherCAT from-slave ready
  • Thermal path to housing
NRB-Sense

Sensor interface boards

Aggregating force, position and environmental channels into one qualified interface, with the analog section EMI-segregated from the digital side.

  • Force/torque conditioning
  • Encoder aggregation
  • EMI-segregated analog section
NRB-Power

Power and battery management

Power distribution and BMS subsystems built around a serviceable pack, so a worn pack is replaced rather than discarded with the robot.

  • Protection and hot-swap
  • Cell-agnostic BMS
  • Field-replaceable architecture
Semiconductor content

What one mid-size platform consumes

Per-unit semiconductor counts for a mid-size humanoid. These are the reference figures we design against, and they explain why the electronics layer holds most of the value in a robot.

Component class
Quantity per unit
Design consequence
MCU
60 – 100
Family choice decides firmware reuse across every module
Power stage
80 – 160
Dominant thermal and cost item
Current sensing channels
60 – 150
Sets the precision ceiling for force control
Encoder interface
40 – 60
Feedback topology has to be fixed early
Gate driver
30 – 70
Isolation requirements differ per platform
Six-axis force sensor
1 – 2
Highest-value and least substitutable part

On these figures: shipment counts, installed base, production volume and order volume differ by roughly 2–6× across sources. Confirm the definition behind any number before relying on it.

Supply constraints

Where the chain is tight

Three places where the electronics layer is limited by something other than design. We plan for these early because none of them are solved by better firmware.

Bottleneck

Planetary roller screws

A recognised bottleneck for linear-actuated joints. Expect lead-time pressure and limited substitution options, which makes joint architecture decisions expensive to revisit.

  • Long lead times
  • Few qualified alternatives
Bottleneck

High-end force sensors

Local manufacturing penetration remains under 30%. This is the single component most likely to constrain a humanoid roadmap, because force control quality depends on it directly.

  • Under 30% local penetration
  • Limits force-control precision
Concentration

Sensor share of BOM

Sensors account for roughly 22% of bill of materials — a share disproportionate to their count, and a useful lens when reading any robot cost breakdown.

  • ~22% of BOM by value
  • Disproportionate to unit count
By platform

What each platform needs most

Household work adds a requirement the industrial platforms never had: the robot lives in an uncontrolled, unstructured environment. That changes what the hardware has to tolerate.

Industrial arms

Repeatability and serviceability dominate. Six to seven joints, a fixed base and a predictable duty cycle.

Repeatable drive behaviourField replacementLong supply life
Quadrupeds

Twelve drive modules and a sensing package that survives outdoors. Weight and cost per joint dominate everything else.

Low mass per jointSealed sensingThermal headroom
AMRs

Duty-cycle endurance in warehouses and hospitals. Safety I/O and power management are the critical paths.

Continuous-duty powerSafety interface clarityFleet serviceability
And humanoids

The hardest case by a wide margin: twenty to twenty-eight modules, force sensing throughout, and a home rather than a controlled lab. This is the platform our training work targets, and it is where the electronics layer has the least room for compromise.

Let's talk about your platform's electronics.

If you are integrating modules into a new platform, the useful conversation starts with the interface — control bus, feedback set, power distribution and thermal envelope. Those four decisions determine most of what follows.

Start a Conversation See Training Programs