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

TL;DR

The paper systematically derives safety constraints for physical human-robot interaction under ISO/TS 15066, highlighting how pain thresholds, tissue stiffness, contact area, and robot/human effective masses shape allowable pre-collision velocity and energy. It advances the field by exposing the impact of modeling assumptions (e.g., fully inelastic vs general restitution) and by contrasting constant ISO masses with configuration-dependent apparent masses, showing substantial safety/performance trade-offs. The work also surveys energy-based safety paradigms, including energy regulation, energy tanks, and safety filters like Control Barrier Functions, while discussing their benefits and limitations, particularly regarding passivity vs. explicit safety guarantees. Practically, it argues for perception-driven, context-aware safety that dynamically adapts limits to task and human state, and it identifies sustained-contact safety as a key open challenge for future standards and controllers. Overall, the paper provides a rigorous framework to interpret safety constraints, quantify their impact on performance, and guide the design of energy-based, perceptive safety architectures in pHRI.

Abstract

This paper aims to provide a clear and rigorous understanding of commonly recognized safety constraints in physical human-robot interaction, i.e. ISO/TS 15066, by examining how they are obtained and which assumptions support them. We clarify the interpretation and practical impact of key simplifying assumptions, show how these modeling choices affect both safety and performance across the system, and indicate specific design parameters that can be adjusted in safety-critical control implementations. Numerical examples are provided to quantify performance degradation induced by common approximations and simplifying design choices. Furthermore, the fundamental role of energy in safety assessment is emphasized, and focused insights are offered on the existing body of work concerning energy-based safety methodologies.

Figures (5)

• Figure 1: Examples of collaborative robots. From left to right: KUKA LBR iiwa, UR5, Doosan H2017, and ABB YuMi (courtesy of KUKA, Universal Robots, Doosan Robotics, and ABB).
• Figure 2: (a) Mechanical model representing a robot-human collision: the moving robot body is modeled as an effective mass $m_R$, the impacted human body region as an effective mass $m_H$, and tissue deformation at the contact point is represented by a spring of stiffness $k$. (b) Simulation results with $m_R = 3.0$ kg, $m_H = 1.0$ kg, $k = 5.0$ N/m, and initial velocity $v_0 = 1.0$ m/s. The plot displays the time evolution of the velocities of the masses $m_R$ and $m_H$, denoted $v_R$ and $v_H$, respectively, as well as the spring compression $\Delta x$. The maximum spring compression occurs after $0.6~\mathrm{s}$, at which point both masses attain approximately the same velocity, $v_* \approx 0.75~\mathrm{m/s}$. The shaded area indicates motion after maximum compression.
• Figure 3: Schematic pipeline for deriving robot motion safety limits in terms of maximum admissible speed $v_{\max}$ and pre-collision kinetic energy $K_{0,\max}$ (red circles) from force- and pressure-based pain thresholds. Solid arrows denote the sequence of analytical derivations, while dashed arrows indicate the dependencies on parameter (blue circles): robot effective mass $m_R$, human effective mass $m_H$, body-region stiffness $k$, and maximum allowable force or pressure before pain $F_{\max}$/$p_{\max}$. The interaction is modeled using the mass-spring-mass representation shown in \ref{['fig:model_mass_spring_mass']}.
• Figure 4: A physical robot–human collision typically occurs in two phases: a transient contact (brief impact governed by robot- and human-reflected dynamics) followed by a quasi-static contact, where the contact can be further subdivided into either pushing (unconstrained) or crushing (clamped) haddadin2016physical.The figure illustrates unconstrained and clamped interaction scenarios. In both cases, the robot approaches with initial velocity $v^R>0$ while the human is initially at rest ($v^H=0$). In the unconstrained case, the human accelerates after impact and eventually recoils, resulting in a pushing interaction with decreasing relative velocity. In the clamped case, the human remains stationary throughout, leading to a crushing interaction.
• Figure 5: Distribution of admissible end-effector velocities across ISO body regions for three safety formulations: transient (T) contact with scaled force/pressure limits (green), quasi-static (QS) contact following ISO/TS 15066 limits (orange), and the potentially-clamped scenario using \ref{['eq:clamped_scenario']} (blue). Boxplots show the variability of velocity limits resulting from the configuration- and direction-dependent apparent mass $m_\mathbf{u}(q)$, while star markers ($\bigstar$) indicate the corresponding scalar limits obtained using the constant ISO-based mass $m_R^{ISO}$.