Annotating Data for Robot Safety Certification

The rapid development of industrial automation and the widespread adoption of collaborative robots (cobots) are significantly transforming modern manufacturing processes. This opens up new opportunities to increase productivity, production flexibility, and operational efficiency.

In modern robotic solutions, data annotation is becoming increasingly important, especially for systems that use computer vision, machine learning, and intelligent decision-making algorithms. High-quality annotated datasets enable robots to recognize human presence, detect potentially dangerous situations, classify operating states, and provide a substantiated assessment of compliance with safety requirements.

Theoretical foundations of the safety of robotic systems and the regulatory framework for certification

Functional safety of robotic systems is a key aspect of modern industrial robotics, determining the system's ability to respond correctly to potentially dangerous situations and prevent harm to humans, equipment, or the environment. In robotic systems, safety is ensured by a combination of hardware solutions, software algorithms, and risk assessment procedures.

The main goal of functional safety is to minimize risks to an acceptable level by implementing protective functions such as emergency stop, speed limitation, human presence monitoring, and control of the robot's workspace.

Classification of industrial and collaborative robots

Industrial robots traditionally operate in isolated environments, separated from humans by physical barriers or protective systems. They are designed to perform high-speed and high-precision operations without direct interaction with the operator.

Collaborative robots (cobots), on the other hand, are designed to work together with a human in the same workspace. Their design and software include built-in safety mechanisms that help prevent injury to the person during the interaction. This is achieved by limiting the force, speed, and energy of the movement, as well as using sensor systems to detect contact or the approach of a person.

Main risks in human-robot interaction

Human interaction with robotic systems is associated with a number of potential risks, including:

mechanical injuries due to collision with the moving parts of the robot.
errors in software or control systems.
incorrect recognition of a person or objects in the work area.
failures of sensor systems.
unpredictable behavior of the robot in a dynamic environment.

Assessment of these risks is a mandatory stage of the development and certification of robotic systems.

Regulatory framework and role of certification

Robotic system certification is based on international standards that define requirements for the design, implementation, and operation of robots with safety in mind. One of the key documents in this area is the ISO 10218 standard, which comprises two parts and sets out requirements for industrial robots and robotic systems.

This standard defines general safety principles, including risk assessment, requirements for control systems, and integration of robots into production environments. Particular attention is paid to the interaction of robots with other machines and operators.

Overview of ISO 10218 and collaborative robot standards

Standard	Title	Main Content	Role in Safety
ISO 10218-1	Robots and robotic devices — Safety requirements for industrial robots	Specifies safety requirements for robot design, built-in protective functions, and motion control systems	Ensures the intrinsic safety of the robot as a standalone machine
ISO 10218-2	Robots and robotic devices — Safety requirements for industrial robots	Defines requirements for integrating robots into industrial environments and interacting with other equipment	Ensures safe deployment and system-level integration
ISO/TS 15066	Robots and robotic devices — Collaborative robots	Defines collaborative operation modes (e.g., speed and separation monitoring, hand guiding) and limits for force and pressure	Ensures safe human–robot interaction in shared workspaces

Key Principles of the Standards

Despite differences in scope, all these standards are based on several common principles:

Risk assessment: identification and evaluation of potential hazards before system deployment
Hierarchical safety measures: prioritization of engineering controls over organizational procedures
Limitation of physical interaction: control of speed, force, and energy during operation
Workspace monitoring: use of sensors to detect human presence
Safe system integration: considering the robot as part of a complete industrial system rather than an isolated device

Importance for Certification

ISO 10218 and ISO/TS 15066 form the foundation for safety certification of industrial and collaborative robotic systems. They define minimum safety requirements that must be satisfied before a system can be deployed in industrial environments.

In modern robotic applications, these standards also serve as a basis for validating perception algorithms, particularly those relying on sensor data and computer vision systems.

Role of Data Annotation in Robot Safety Certification

Data annotation is the process of labeling raw sensor data (such as images, video streams, point clouds, or time-series signals) with meaningful information for machine learning and system validation. In robotics, annotated datasets are a foundational component for developing perception systems that enable robots to interpret their environment and make safety-critical decisions.

In the context of industrial and collaborative robots, annotation typically focuses on identifying humans, objects, motion patterns, and potentially hazardous situations within the robot’s operational workspace.

Types of Annotations Relevant to Safety Systems

Several types of annotation are commonly used in robotic safety-related applications:

Object detection annotations: bounding boxes around humans, tools, or machinery.
Semantic segmentation: pixel-level labeling of safety-critical regions (e.g., human body parts, robot workspace zones).
Pose estimation labels: "keypoints" representing human posture and movement.
Event-based annotations: labeling of dynamic safety events such as collisions, near-miss situations, or unsafe proximity.
3D annotations: spatial labeling using LiDAR or depth data for accurate distance estimation.

Importance of Annotation Quality for Safety Certification

In safety certification processes based on standards such as ISO 10218, system behavior must be validated under a wide range of operational scenarios. When machine learning components are involved, the reliability of these systems strongly depends on the quality of training and validation data.

Poor or inconsistent annotation can lead to:

incorrect detection of human presence.
failure to recognize hazardous situations.
false positives in emergency stopping systems.
reduced reliability of risk assessment models.

Methods and tools for data annotation

Method	Description	Advantages	Limitations
Manual annotation	Human experts label data frame-by-frame or point-by-point	High accuracy, strong domain understanding	Time-consuming, expensive, not scalable
Semi-automatic annotation	AI pre-labels data, humans refine results	Faster than manual annotation, improved consistency	Requires model training, still needs human oversight
Automatic annotation	Fully automated labeling using pretrained models	Highly scalable and fast processing	Lower accuracy in complex or safety-critical scenarios
Quality assurance (QA) workflows	Review and correction of annotated datasets	Improves reliability and consistency of labels	Requires additional time and resources
Annotation quality metrics	Evaluation of labeling accuracy and consistency	Enables objective quality control	Depends on availability of ground truth or reference data
Data management tools	Organization, versioning, and tracking of datasets	Ensures traceability and reproducibility	Does not perform annotation itself

FAQ

What is a cobot safety dataset?

A cobot safety dataset is a structured collection of labeled data used to train and validate safety systems in collaborative robots. It typically includes images, depth data, or sensor streams showing human-robot interaction scenarios. The goal is to support safe decision-making in shared workspaces.

Why is ISO 10218 training data important for robotics safety?

ISO 10218 training data ensures that robotic systems are evaluated against standardized safety requirements. It helps validate whether robots behave safely in industrial environments. Such datasets are essential for certification and compliance processes.

What is collaborative robot annotation?

Collaborative robot annotation refers to labeling data that represents interactions between humans and cobots. This includes identifying human presence, robot motion, and interaction zones. It is crucial for training perception models in safety-critical systems.

How does human-robot interaction safety depend on data?

Human-robot interaction safety depends on accurate perception of human location and behavior. Annotated datasets help robots recognize risky situations in real time. Poor data quality can lead to unsafe decisions or system failures.

What is force limitation labeling in cobot systems?

Force-limitation labeling defines thresholds for acceptable contact force between robots and humans. It is used to train systems that prevent injuries during physical interaction. This directly supports compliance with collaborative safety standards.

What is risk zone segmentation in robotics?

Risk zone segmentation divides the workspace into safe, warning, and dangerous regions. These zones help robots adjust speed or stop when humans enter critical areas. It is a key component of safety-aware perception systems.

How is annotation used in ISO 10218 compliance?

Annotation is used to create datasets that simulate real industrial scenarios. These datasets help verify whether robotic systems meet ISO 10218 safety requirements. They support testing of detection, tracking, and risk assessment models.

What challenges exist in collaborative robot annotation?

Key challenges include ambiguity in labeling safety boundaries and variability of industrial environments. Another issue is the need for high precision in dynamic human-robot interactions. Expert knowledge is often required for reliable annotation.

Why is dataset quality critical for cobot safety?

High-quality datasets ensure reliable perception and decision-making in cobots. Errors in annotation can lead to false detection or missed hazards. This directly impacts the safety of human-robot collaboration.

How does risk labeling improve robot behavior?

Risk labeling helps robots distinguish between safe and unsafe conditions. It allows adaptive behavior such as slowing down or stopping near humans. This improves overall safety in collaborative environments.