See the interfaces among data, control, hardware and people.
Robotics technology stack · Layer 01
Perception and sensor fusion
How cameras, depth, lidar, force, tactile and proprioceptive signals become a time-aligned estimate a robot can safely act on.
Evidence statusEstablished methods; deployment remains environment-specificLast reviewed: 11 August 2026
01
FUURAA thesis
More sensors do not automatically create better perception; calibration, timing, uncertainty and failure detection determine whether fusion is trustworthy.
Reading method
Turn a technology label into an engineering chain, testable claims and explicit boundaries.
Translate capability into task, latency, failure and recovery.
Separate standards, independent measurement, research and first-party claims.
State what cannot be inferred from a demo, benchmark or interface.
System breakdown
Four interdependent layers determine whether the technology can enter real work.
Each layer shows its role and the failure signal most worth watching.
Sensor geometry
Intrinsic and extrinsic calibration locate measurements in a shared frame.
Time and transport
Timestamps, latency budgets and dropped-message handling determine whether observations describe the same physical moment.
State estimation
Filtering and learned estimators combine noisy evidence while preserving confidence and observability.
Fault response
Plausibility checks, redundancy and degraded modes keep one bad stream from silently corrupting control.
Engineering evaluation
Five checks turn abstract capability into reviewable system evidence.
Record normal performance, failure, recovery and human cost—not only the best-looking result.
- 01
Test the operating envelope
Measure by lighting, weather, dust, occlusion, range, speed and surface—not one aggregate score.
- 02
Record ground truth
Use traceable targets and synchronized reference measurements for detection, tracking and pose error.
- 03
Measure uncertainty quality
A confidence estimate should predict error frequency, not merely look numerically precise.
- 04
Inject sensor faults
Disconnect, delay, saturate and miscalibrate inputs to verify detection and degraded operation.
- 05
Connect perception to risk
Translate misses and late detections into stopping distance, collision exposure and task consequences.
Scope boundaries
State what the evidence supports—and what it does not.
- 01
A benchmark dataset cannot establish performance in an unseen site or sensor configuration.
- 02
Semantic recognition is not a certified protective function unless the complete safety chain is validated for that use.
- 03
Redundant sensors may share common-cause failures such as contamination, time drift or power loss.
Sources and evidence status
Keep the source, date, evidence identity and reading boundary visible.
This page prioritises standards bodies, public measurement programmes, official project documentation and original research disclosures.
01
NIST · 2014-07-30
↗02Performance Metrics for Object and Human Detection and Tracking Systems
Defines comparable detection, tracking, intrusion and pose metrics using common tasks and ground truth.
NIST · 2016-01-22
↗Performance Evaluation of Human Detection Systems for Robot Safety
Connects quantitative human detection and tracking measures to collaborative robot safety decisions.