Physical Human-Robot Interaction: A Critical Review of Safety Constraints

Authors: Riccardo Zanella, Federico Califano, Stefano Stramigioli Affiliation: Robotics & Mechatronics (RaM) Group, University of Twente, The Netherlands arXiv: 2601.19462 (v3, 2026-04-02, eess.SY) — full text fetched from https://arxiv.org/html/2601.19462v3

Overview

Examines commonly recognized safety constraints in physical human-robot interaction, particularly ISO/TS 15066. Critically investigates how these constraints are derived, questions foundational assumptions, and assesses real-world applicability in manufacturing.

ISO/TS 15066 and Collaborative Robotics

ISO/TS 15066 defines four collaborative operation modes:

• Speed and Separation Monitoring (SSM)
• Hand Guiding
• Safety-Rated Monitored Stop
• Power and Force Limiting (PFL) — the mode involving direct physical contact

PFL mode sets quantitative limits on contact force and pressure to prevent injury, enabling humans and robots to share workspace.

Key Findings

• Energy is fundamental to safety assessment — many existing standards underspecify energy-based analysis
• Common design approximations introduce performance losses that are rarely quantified
• Foundational assumptions in ISO/TS 15066 are not always validated against real-world contact dynamics
• Standards shift (ISO 10218-2 2025 revision): collaborative designation belongs to the application, not the robot itself

Implications

Critical safety review has practical implications for cobot deployments in manufacturing — over-conservatism (from poorly derived constraints) reduces productivity, while under-conservatism creates injury risk.

Full Text (arXiv v3, 2026-04-02)

Abstract

This paper provides rigorous analysis of safety constraints in physical human-robot interaction, particularly regarding ISO/TS 15066 standards. The authors investigate constraint derivations, examine underlying assumptions, and evaluate practical implications for system safety and performance in industrial contexts. Key design parameters within safety-critical control architectures are identified with numerical examples quantifying performance degradation. The analysis emphasizes energy’s fundamental role in safety assessment and provides insights into energy-based methodologies for collaborative industrial robot systems.

1. Introduction

Collaborative robots represent a paradigm shift in industrial automation, enabling direct physical interaction with human operators in shared workspaces. Human-robot collaboration can reduce cycle times by up to 50% compared to all-human operations. Physical Human-Robot Interaction (pHRI) encompasses all scenarios involving physical contact, where forces, energy, and power exchanged at the interface are critical quantitative metrics. Safety is context-dependent and emergent, influenced by robot behavior, environment, task requirements, and human actions.

ISO/TS 15066 (2016) introduced technical specifications supporting pHRI safety through quantitative force, velocity, acceleration, and energy transfer limits. Cobots’ inherent design creates misconceptions about automatic safety, potentially leading to inadequate control measures — the 2025 revision of ISO 10218-2 clarified that only applications, not robots themselves, can be designed as collaborative.

Five primary contributions:

1. Systematic derivation of safety constraints within the Power and Force Limiting (PFL) mode of ISO/TS 15066
2. Critical identification and justification of underlying assumptions
3. Analysis of constraint implications for both safety and robotic efficiency
4. Concise review of energy’s role in safety assessment and energy-based control strategies
5. Identification of key design choices affecting safety-performance trade-offs with quantitative investigation

2. Operational Constraints of Robotic Motion

ISO/TS 15066 represents human safety through maximum allowable force and pressure values per body region, derived from the FP-0317 Project (University of Mainz, quasi-static pain-onset thresholds across 29 anatomical regions in 100 healthy adults) and DGUV/BGIA recommendations. These are quasi-static thresholds; ISO/TS 15066 extends them to transient contacts (<0.5s) via scaling factors — allowable force/pressure for transient contact is set to at least twice the quasi-static value.

Mass-spring-mass model: the collision is idealized as two lumped masses (robot effective mass m_R, human effective mass m_H) connected by a linear elastic spring (stiffness k). Conservation of momentum and energy yield three equivalent operational constraints:

1. ΔK ≤ U_{s,max} (energy constraint)
2. K₀ ≤ K_{0,max} (kinetic energy constraint)
3. v₀ ≤ v_{0,max} (velocity constraint)

where v_{0,max} = √[(m_R+m_H)/(m_R·m_H)] × F_max/√k.

3. Remarks on Deriving Robot Motion Constraints

• Inelastic assumption: ISO/TS 15066 assumes fully inelastic collision (all lost kinetic energy dissipates through tissue). This is a strong, not universally representative assumption — real impacts sit between perfectly elastic and inelastic.
• Pain thresholds: max allowable speed scales linearly with F_max and inversely with stiffness k and contact area; tabulated F_max values assume a conservative 1cm² contact area.
• Human effective mass: the “clamped” case (m_H → ∞, body segment immobilized and unable to recoil) is more conservative than the general case where the operator can recoil. ISO/TS 15066 does not explicitly separate these cases — the paper argues it should. Example: forearm anatomical mass ~1.4kg but effective mass 2kg; thigh anatomical mass ~9.8kg but effective mass 75kg (whole-body involvement when standing).
• Robot effective mass: ISO’s quasi-static approximation m_R^ISO = M_total/2 + m_l is often overly conservative. A dynamic, configuration-dependent formulation m_u(q) = (uᵀΛ_ν⁻¹(q)u)⁻¹ gives a more accurate estimate along the actual contact direction and can be exploited via kinematic redundancy to reduce injury risk while improving performance.

4. Energy in Safe pHRI

An alternative safety framing uses maximum mechanical energy absorbed by the human body before irreversible damage (drawing on automotive safety / Head Impact Power literature) rather than instantaneous peak force. Energy-based control strategies surveyed: impedance control with stiffness modulation, port-Hamiltonian frameworks, optimization-based approaches, virtual energy tanks, Control Barrier Functions (CBFs), power regulation (stiffness limits energy, damping limits power), valve-based energy tanks, and learning-based (RL) controllers that implicitly preserve passivity.

Key caution: passivity ≠ safety. A passive system with large (but bounded) stored energy can still produce dangerous power ejections over short intervals. The paper also flags that a common simplified constraint (K₀ ≤ U_{s,max}) implicitly assumes the clamped/infinite-effective-mass case and is therefore an overly conservative special case, not the general rule.

5. Final Discussion + Quantitative Example

Safety limits should ideally adapt dynamically to which body region is at risk and whether the human can recoil — but most industrial setups lack the perception maturity for this and instead assume worst case (clamped scenario, most conservative body region, typically the face), which is safe but costly to productivity.

Quantitative example (Franka Emika Panda, ISO quasi-static m_R^ISO = 5.545 kg): the clamped-scenario speed limits are lower than the ISO quasi-static formulation across body regions — largest discrepancies for neck, hands/fingers, and arms (low tissue stiffness, small effective mass). Using the configuration-dependent effective mass m_u(q) instead of the constant ISO value yields consistently higher (less conservative, still safe) speed limits.

6. Conclusion and Future Research Directions

Energy is the fundamental quantity linking safety requirements to operational constraints. Future directions: advanced perception for body-region/clamped-scenario detection enabling dynamic limit adaptation; refined real-time effective-mass computation under variable payload/compliance/friction; updated empirical pain/injury thresholds from controlled impact studies; integrated safety-performance optimization; formal uncertainty quantification for safety constraint derivation.