State
Measure estimator error under slip, impact, darkness, dust and sensor disagreement.
System anatomy · 05
Legged mobility and whole-body control
Evaluate perception, state estimation, foothold selection, balance and recovery as one coupled control loop.
Evidence statusField mission framework · task, terrain and site validation requiredLast reviewed: 15 August 2026
05
FUURAA thesis
A strong nominal gait is not robust locomotion unless uncertainty and disturbance are tested.
Engineering map
Turn the title into engineering objects that can be observed, measured and reviewed.
Each layer states the boundary to establish and the evidence needed for the next decision.
Contact
Test foothold geometry, friction, compliance and negative obstacles.
Recovery
Demonstrate stop, stabilisation, self-righting or assisted retrieval after loss of balance.
Verification questions
Write the questions first, then decide whether a demo, test, pilot or operating record can answer them.
Each question needs an object, conditions, denominator, threshold and accountable decision owner.
- 01
Which disturbance defines failure?
- 02
How does payload shift the stability margin?
- 03
What happens after estimator disagreement?
- 04
Can recovery create a new hazard?
Evidence to preserve
Enable the next reader to reconstruct conditions, results, failures and the decision.
A conclusion alone loses reviewability; raw records, configuration and exclusions matter too.
- 01
Evidence package 1
Repeated terrain score distribution
- 02
Evidence package 2
Estimator and contact fault log
- 03
Evidence package 3
Fall and recovery demonstration
Scope boundary
State what this evidence still cannot be generalised to.
- 01
Simulation, domain randomisation and a single real-robot run do not exhaust long-tail controller failures.
- 02
This is an engineering reading and decision framework; it does not replace application-specific risk assessment, conformity work, procurement acceptance or professional advice.
Sources and evidence status
Read standards scope, measurement evidence and application conclusions separately.
Source dates and review status remain visible; external sources open in a new tab.
01
NIST · Updated 2022-02-25 · verified 2026-08-15
↗02Legged Mobility and Search Tests
Uses reconfigurable grating, hills, holes and search targets to measure locomotion and inspection together under increasing difficulty.
Robotics: Science and Systems · Published 2024-07
↗03Rethinking Robustness Assessment for Learning-based Quadrupedal Controllers
Shows that low-magnitude sequential disturbances can expose failures in robust locomotion policies and that domain randomisation alone is insufficient.
Unitree Robotics · Manufacturer page · verified 2026-08-15
↗Unitree B2 industrial quadruped specifications
Publishes configuration-dependent figures for speed, load, endurance, stairs, environmental rating, sensors and development interfaces.