Physical AI & Spatial Intelligence Glossary

GLOSSARY TERMS

3

3D Perception

3D perception is the ability of an AI system to detect, locate, and classify objects in three-dimensional space in real time. Unlike 2D image analysis, which infers depth from visual cues, 3D perception uses direct spatial measurements from sensors such as LiDAR or stereo cameras to produce an accurate three-dimensional understanding of a scene. 3D perception enables more reliable object classification (distinguishing a person from a vehicle, or a person from a tree branch), more accurate position and trajectory measurement, and more robust performance under challenging conditions such as darkness, shadows, or adverse weather.

---

A

Autonomous Guided Vehicle (AGV)

An autonomous guided vehicle (AGV) is a mobile robot that navigates a defined environment without a human operator, typically used in warehouses, factories, ports, and distribution centers to move materials. AGVs rely on perception systems, including LiDAR, to detect their environment, avoid obstacles, and navigate safely alongside human workers. LiDAR-based safety zones and proximity detection systems mounted on AGVs or on facility infrastructure protect workers by triggering slowdowns or stops when people enter defined zones around moving vehicles.

Autonomous Sensing

Autonomous sensing is the continuous, self-directed collection and interpretation of sensor data by an AI system without requiring human instruction for each measurement cycle. An autonomous sensing system detects, classifies, and tracks objects in real time, updates its world model continuously, and generates alerts or outputs whenever defined conditions are met. LiDAR-based autonomous sensing systems from Quanergy operate 24/7 without requiring an operator to watch a video feed, generating alerts only when events of interest are detected.

---

C

Collision Avoidance

Collision avoidance is the capability of a physical AI system to detect objects or people in the path of a moving vehicle or machine and trigger a response to prevent impact, such as slowing down, stopping, or redirecting the machine. LiDAR-based collision avoidance systems scan the environment in three dimensions at high update rates, detecting obstacles reliably in all lighting conditions and without requiring reflectors or targets to be placed in the environment. Collision avoidance is a critical safety function in AGV fleets, port equipment, mining vehicles, and construction machinery.

---

D

Digital Twin

A digital twin is a real-time virtual representation of a physical environment that mirrors the state and activity of that environment continuously from live sensor data. In physical AI applications, a digital twin built from 3D LiDAR data shows the positions, movements, and classifications of all detected objects within a facility or space, updating in real time as conditions change. Digital twins enable operators to monitor complex environments from a central dashboard, replay historical events for incident investigation, simulate the impact of layout or operational changes before implementation, and develop and test AI models against realistic data.

---

E

Edge Inference

Edge inference is the execution of AI algorithms directly on or near a sensor, at the point of data collection, rather than transmitting raw data to a remote server for processing. In LiDAR systems, edge inference means that object detection, classification, and tracking models run on the sensor itself or on a co-located compute module, producing detection events rather than raw point clouds. Edge inference reduces latency to under 100 milliseconds, reduces bandwidth requirements by orders of magnitude, eliminates dependency on network connectivity for core detection functions, and improves data privacy by avoiding transmission of raw spatial data offsite. Quanergy Q-SHIELD supports on-sensor zone configuration, allowing detection zones and classification rules to be defined and executed directly on the sensor without a central processing dependency.

---

G

Geofencing (Physical AI)

Geofencing in physical AI is the definition of virtual boundaries within a real-world space that trigger automated responses when objects enter, exit, or linger within them. Unlike GPS-based geofencing used for mobile devices, LiDAR-based geofencing defines boundaries in three-dimensional space at sub-meter precision. When a person or vehicle crosses a defined boundary, the system generates an immediate alert or triggers an automated response such as activating a camera, initiating an alarm sequence, or sending a command to a control system. Quanergy Q-TRACK and Q-TRACK support multi-zone geofencing with configurable alert escalation sequences.

---

M

Multi-Object Tracking (MOT)

Multi-object tracking (MOT) is the simultaneous tracking of multiple distinct objects within a sensor’s field of view, maintaining a unique identifier for each object across consecutive measurement frames. MOT algorithms must handle objects entering and leaving the scene, objects occluding each other temporarily, and objects moving at different speeds and in different directions. In LiDAR-based systems, MOT is performed on 3D point cloud data, tracking each object as a bounding box with a persistent track ID, velocity vector, and classification label. Accurate MOT is foundational to all downstream spatial analytics, including trajectory analysis, dwell time measurement, and crowd density calculation.

---

O

Object Trajectory

An object trajectory is the path that a tracked object follows through space over time, represented as a sequence of positions at successive time steps. Trajectory data enables analysis of movement patterns, identification of common routes through a space, detection of unusual or anomalous movements, and prediction of where an object is heading. In perimeter security, trajectory analysis can distinguish objects moving toward a protected asset from objects moving parallel to a boundary. In crowd analytics, trajectory data reveals the paths passengers most commonly follow through a terminal, informing wayfinding and retail placement decisions.

Occupancy Grid

An occupancy grid is a spatial representation of an environment that divides the space into a regular grid of cells, with each cell labeled as occupied, free, or unknown based on sensor measurements. Occupancy grids are used in robotic navigation and physical AI to represent where objects are and where open space exists for movement. 3D LiDAR produces occupancy grids with high accuracy and at high update rates, enabling real-time navigation for AGVs and robots and real-time crowd density mapping for airport and venue applications. The resolution of the occupancy grid (cell size) determines the precision of the spatial representation.

---

P

Persistent Tracking

Persistent tracking is the maintenance of a consistent track identity for an object across multiple sensors, sensor handoff zones, or temporary occlusion events over an extended time period. Where standard object tracking may assign a new ID to an object that was momentarily lost from view, persistent tracking uses position prediction, shape matching, and motion signatures to reconnect the track and maintain continuity. Persistent tracking is important in large-area security monitoring where a single sensor does not cover the full environment, and in airport analytics where passengers must be followed from check-in through multiple processing stages.

Physical AI

Physical AI is artificial intelligence that perceives, interprets, and acts within the physical world using real-time sensor data from the environment. Physical AI systems combine sensors such as 3D LiDAR, cameras, radar, and IMUs with AI models for detection, classification, tracking, and decision-making to build a continuous, dynamic understanding of their physical surroundings. Unlike AI applied to static datasets or language tasks, physical AI operates in real time and must handle the unpredictability, scale, and environmental complexity of the real world. Quanergy develops physical AI solutions using 3D LiDAR and edge AI to deliver spatial intelligence for security, safety, logistics, and people flow applications at scale.

Proximity Detection

Proximity detection is the real-time identification of when a person, vehicle, or object comes within a defined distance of another object, asset, or zone boundary. In industrial and warehouse environments, proximity detection protects workers from moving equipment by triggering alerts or machine slowdowns when a person enters the operational radius of a robot or vehicle. LiDAR-based proximity detection operates in three dimensions with sub-meter accuracy, detecting people reliably regardless of lighting conditions and without requiring workers to carry tags or devices.

---

R

Real-Time Analytics

Real-time analytics processes and delivers insights from data within milliseconds to seconds of the data being collected, enabling immediate decision-making and automated response rather than post-hoc analysis. In physical AI, real-time analytics transforms raw LiDAR point cloud data into actionable metrics such as occupancy counts, queue lengths, crowd density, and security alerts, available in live dashboards and via API integrations. Real-time analytics is distinct from historical analytics, which aggregates data over time for trend analysis and planning. Both are delivered by Quanergy Q-INSIGHTS: real-time for operational decisions and historical for planning and reporting.

Real-Time Event Detection

Real-time event detection identifies specific occurrences of interest within a sensor’s field of view as they happen, triggering immediate alerts or automated responses. Examples of events detected in LiDAR-based physical AI systems include a person crossing a virtual tripwire, a vehicle entering a sterile zone, an object being left unattended, a person falling, a crowd density threshold being exceeded, or a vehicle dwelling too long in a restricted area. Real-time event detection is the primary mechanism through which physical AI systems convert continuous sensor streams into actionable security and safety outputs.

Re-identification (ReID)

Re-identification (ReID) is the problem of recognizing that an object or person seen by a sensor at one point in time or space is the same object or person seen previously, after a gap in tracking. ReID is necessary when a person passes out of one sensor’s coverage area and into another, or when tracking is briefly lost due to occlusion. In camera-based systems, ReID uses visual appearance features. In LiDAR-based systems, ReID uses 3D shape features, size measurements, and motion signature matching. Anonymous LiDAR-based ReID can maintain track continuity without capturing or storing any biometric information.

---

S

Safety Zone

A safety zone is a defined three-dimensional volume of space around a machine, vehicle, or hazard within which the presence of a person triggers a protective response. Safety zones are a fundamental concept in industrial and warehouse safety, enforced either by physical barriers or by sensor-based detection systems. LiDAR-based safety zones detect person presence within the defined volume in real time and trigger tiered responses: a warning zone at greater distance triggers an audible alarm and machine slowdown, while an inner stop zone triggers an immediate machine halt. This tiered approach reduces false stops while maximizing worker protection.

Scene Understanding

Scene understanding is the capability of an AI system to interpret not just what objects are present in a space, but the relationships between them, their states, and the overall context of the scene. A scene-understanding system does not simply detect a person and a vehicle separately; it understands that the person is walking toward the vehicle, that both are in a restricted zone, and that this combination of conditions may constitute a security event. Scene understanding in physical AI requires accurate 3D perception, object classification, multi-object tracking, and reasoning about spatial relationships and temporal context.

Semantic Map

A semantic map is a spatial representation of an environment that encodes not only the geometry of the space but also the meaning or function of each area. In physical AI, a semantic map might label zones as a pedestrian walkway, a vehicle lane, a security perimeter, a retail area, or a loading dock. Semantic maps allow AI systems to apply context-specific rules: a person detected in a vehicle lane triggers a different response than a person detected in a pedestrian area. Semantic maps are created by overlaying zone labels and business logic onto the base 3D spatial model of the environment.

Sensor Fusion

Sensor fusion is the integration of data from two or more different sensor types to produce a more accurate, complete, or robust understanding of an environment than any single sensor could provide. In physical AI applications, LiDAR is commonly fused with cameras to add color and appearance information to 3D detections, with radar to extend range and improve performance in rain or fog, or with access control and identity systems to add behavioral and identity context to spatial detections. Quanergy Q-INSIGHTS is designed for integration with leading VMS platforms including Milestone, Genetec, Axis, Hanwha, and Bosch, enabling multi-sensor deployments where LiDAR handles detection and classification while cameras provide visual verification.

Situational Awareness

Situational awareness is the comprehensive, real-time understanding of an environment, including what objects and people are present, where they are, what they are doing, how conditions are changing, and what events may be developing. In physical security, situational awareness means that an operator or an automated system has a complete, current picture of activity across a protected area. 3D LiDAR spatial intelligence improves situational awareness by providing accurate, real-time data on the position and movement of all objects in the monitored space, including in darkness and adverse weather conditions where human visual monitoring is unreliable.

Spatial Data

Spatial data is any data that describes the location, size, shape, or movement of objects or features in physical space. In the context of LiDAR-based physical AI, spatial data includes 3D point clouds, object bounding boxes, trajectory paths, occupancy grids, density maps, and zone-level aggregations. Spatial data is the raw material from which all spatial intelligence is derived. The quality, accuracy, and update rate of spatial data directly determines the performance of the physical AI systems built on top of it.

Spatial Intelligence

Spatial intelligence is the capability of an AI system to understand, represent, and reason about the three-dimensional physical world in real time. A spatially intelligent system knows where objects are, what they are, how they are moving, and what is likely to happen next. Spatial intelligence is the core capability of physical AI applications in security, safety, logistics, and people flow analytics. Quanergy delivers spatial intelligence through its 3D LiDAR sensing platforms and the Q-TRACK and Q-INSIGHTS software, enabling customers to build real-time, AI-powered awareness of their physical environments at scale.

---

T

Track Handoff

Track handoff is the process of transferring the tracking responsibility for an object from one sensor to another as the object moves from the coverage zone of the first sensor into the coverage zone of the second. Successful track handoff maintains the same track ID and accumulated history for the object across sensors, enabling end-to-end tracking of a person or vehicle through a large facility without losing continuity at sensor boundaries. Track handoff requires that adjacent sensors’ coverage zones overlap and that the handoff algorithm matches objects between the two streams in real time based on position, velocity, and classification.

Track ID

A track ID is the unique identifier assigned to a detected object at the moment the system first detects it, and maintained for as long as the object remains within the sensor’s field of view (or across sensors in a multi-sensor deployment with persistent tracking). The track ID allows all data points associated with a single object, including position history, velocity, classification, and zone interactions, to be linked and analyzed as a unified record. In anonymous tracking systems, the track ID is a numerical identifier only, with no linkage to personal identity data.

---