Active and Compliant Safety in Shared Human-Robot Workspaces

74 concepts67 papers2 proposals

Traditional industrial automation relies on a simple safety paradigm: physical separation. Robots operate inside rigid metal cages, and if a human enters the workspace, the system triggers an emergency stop. However, as robots transition into unstructured human environments—ranging from collaborative manufacturing lines to domestic care—this binary approach fails. Halting a robot every time a human draws near paralyzes productivity and prevents fluid, close-proximity collaboration.

To allow humans and robots to safely coexist, researchers are shifting from rigid isolation to a dual-layered paradigm: predictive algorithmic safety shields that dynamically adjust to human movement, and inherently compliant hardware that minimizes the physical consequences of accidental contact.

Dynamic and Predictive Safety Boundaries

The core limitation of traditional safety systems is that they treat humans as static obstacles. To maintain fluid interaction, control frameworks must treat humans as active, moving participants whose future states can be modeled and anticipated.

A powerful mathematical tool for enforcing these boundaries is the Control Barrier Function (CBF). A CBF defines a set of safe states (e.g., a minimum separation distance) and mathematically guarantees that the robot's control inputs will never allow the system to leave this safe set.

Standard Safety Zone:   [ Robot ] -------- Fixed Radius -------- | Human (Triggers Stop)

Dynamic Safety Zone:     [ Robot ] <--- Relative Velocity ---> [ Human (Adjusts Fluidly) ]

While early applications of CBFs focused on static spatial boundaries, recent work adapts them to the fluid dynamics of human movement. For instance, Aaron Ames demonstrates how CBFs can define variable assistance boundaries in lower-body exoskeletons, ensuring user stability by dynamically constraining movements within a safe operating envelope.

To make these boundaries even more responsive, Zihan Zeng et al. revised the standard CBF formulation to explicitly incorporate the relative velocity between humans and robots. The intuition is straightforward: a human standing still poses a different level of threat than a human running toward the machine. By factoring in relative velocity, the safety shield expands when closing speeds are high and contracts when movements are cautious, preventing unnecessary stops while preserving absolute safety guarantees.

Complementing these formal control boundaries is human-aware motion planning. Rather than reacting only when a boundary is violated, robots can proactively plan trajectories that avoid human workspaces entirely. Xuan Zhao et al. integrate human occupancy maps into a stochastic trajectory optimizer. By predicting where a human co-worker is likely to move next based on past patterns, the robot shapes its path to minimize both physical risk and the psychological stress of close-proximity interaction.

Hardware Compliance and Kinetic Energy Management

Algorithms alone cannot guarantee safety if the physical hardware is inherently dangerous. High-mass rigid robots operating at high velocities possess immense kinetic energy. In a brief collision, this energy is transferred almost instantly, generating lethal peak forces. As Alessandra Rossi notes, a 40kg humanoid robot moving at athletic speeds can generate up to 16 kN of force during an impact—highlighting the critical need for advanced kinetic energy management.

To mitigate this risk, researchers leverage two hardware strategies: reducing moving mass and embedding mechanical compliance.

Rigid Actuation:    [ Heavy Motor + Rigid Link ] ===> High Inertia ===> High Impact Force

Tendon-Driven:      [ Heavy Motor at Base ] --- Tendon --- [ Light Link ] ===> Low Impact Force

Tendon-driven architectures achieve this by relocating heavy actuators away from the moving joints down to the robot's base. By minimizing the arm's mass, the robot's reflected inertia drops drastically. Simon Guist et al. utilize this design in the PAMY2 robot arm, combining a tendon-driven structure with pneumatic artificial muscles to achieve high-speed dynamic motions that remain safe for human interaction. Similarly, Yuze Wu et al. employ an underactuated tendon mechanism to create a lightweight, hand-sized flying robot capable of safe indoor aerial manipulation near people.

When contact does occur, embedded compliance acts as a mechanical buffer. Soft and inflatable robotic structures increase the duration of an impact, which fundamentally lowers the peak force experienced by the human body. Matthias Hofer et al. demonstrate this with an inflatable spherical robotic arm that maintains robust manipulation capabilities while remaining completely safe during unexpected physical contact.

For end-effectors, integrating passive compliance—such as spring-loaded linear sliders—provides a critical margin for error. Guilherme Maeda et al. show that this mechanical buffer significantly reduces peak contact forces during grasping, protecting both the robot and the human environment without requiring ultra-high-precision force sensors.

Force Regulation and Intentional Contact

In close-range assistive tasks, such as robot-assisted dressing or physical therapy, contact is not an accident to be avoided—it is a functional requirement. The challenge then shifts from collision avoidance to precise force regulation and contact classification.

Binary Stop:         [ Any Contact Detected ] --------> Emergency Stop (Disrupts Workflow)

Intelligent Contact: [ Contact Detected ] -- DML Classifier --> Intentional Touch (Continue)
                                           \--> Accidental Collision (Safely Yield)

When a robot must directly manipulate or interact with the human body, it requires multi-modal sensing to ensure forces remain well below injury thresholds. Zhanyi Sun et al. address this with the Force-Constrained Visual Policy (FCVP) framework for robot-assisted dressing. By pairing a vision policy trained in simulation with a real-world force-dynamics model, the system successfully manipulates deformable garments around a user while strictly limiting the physical pressure applied to the human body. This delicate touch is echoed in soft robotic end-effectors designed by Uksang Yoo et al., which utilize tendon tension sensing and structural compliance to perform close-range hair care safely and comfortably.

Finally, true collaboration requires the robot to understand the meaning of touch. Standard safety protocols treat all physical impacts identically, triggering a shutdown whether the contact was an accidental bump or a deliberate guiding touch.

To break this limitation, Maryam Rezayati developed a Deep Metric Learning (DML) approach that maps robot sensor data into a specialized similarity space. This framework allows the robot to distinguish between intentional human touch and accidental collisions with high accuracy. By recognizing when a human is intentionally guiding it, the robot can transition to an active impedance control mode—adjusting its stiffness and damping in real time (Loris Roveda et al.)—to facilitate collaboration rather than shutting down.

Key Foundational Work

1P1-F4 Development of the New Motor Driver and Introduction of the Products in Collaboration (2001) — contributed 1 concept elements to this direction.
Development of a Reasonable Force Sensor for a Standing-up and Sitting Motion Support System (2004) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_human_robot_interaction_safety_b5f917b9,
  title = {human-robot interaction safety},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/human-robot-interaction-safety/rdr_b5f917b937484f709b840394cbeac53b},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_human_robot_interaction_safety_b5f917b9_v1,
  title = {human-robot interaction safety},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/human-robot-interaction-safety/rdr_b5f917b937484f709b840394cbeac53b/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake