Safety Centric Interaction for Collaborative Robotics

This direction is no longer present in the current build. Its last version stays readable and citable here.

Version 1 of 173 papers86 conceptsfrozen 2026-10-04

For decades, industrial safety was synonymous with segregation. To keep humans safe, we placed robots behind steel cages or static light curtains. However, as robotics moves from the sterile predictability of factory floors to the chaotic reality of construction sites, homes, and shared warehouses, the "cage" model has become a bottleneck.

The emerging research direction in Safety-Centric Interaction seeks to replace physical barriers with dynamic, multi-layered safety protocols. This shift moves safety from a hardware constraint to an active, intelligent behavior, allowing robots to work in close proximity to humans without sacrificing productivity or well-being.

From Static Fences to Dynamic Envelopes

The primary challenge in fenceless collaboration is the "impedance mismatch" between human unpredictability and robotic rigidity. Traditional systems often trigger an emergency stop the moment a human enters a predefined zone, which Zhe Huang notes can render a robot inefficient in high-mix production environments.

To solve this, researchers are developing "virtual safety envelopes" that move with the robot. Karnik Ram and colleagues introduced Programmable Light Curtains (PLCs) that project dynamic safety boundaries in real-time. Unlike traditional LiDAR, which requires massive computational overhead to process 3D point clouds, these curtains focus only on the relevant volume of space, enabling high-speed monitoring at a fraction of the cost.

Complementing this, Claus Lenz demonstrated how fusing multiple Kinect sensors can create a redundant 3D model of the workspace. By implementing "robot suppression"—filtering the robot’s own body out of the sensor data—the system can distinguish between the machine's intended motion and an encroaching human limb.

Decoupling Efficiency from Reflexes

A key architectural insight in this field is the separation of high-level planning from low-level safety. Changliu Liu proposed the Robot Safe Interaction System (RSIS), which uses a parallel planning architecture. In this model:

  1. Global Planning focuses on long-term efficiency and task completion.
  2. Local Reflexes operate on a much faster loop, overriding the global plan to avoid collisions.

This hierarchy allows the robot to remain proactive. For instance, Liting Sun developed a framework that combines plan recognition with trajectory prediction. By understanding why a human is moving toward a certain tool, the robot can adjust its path before a collision is even imminent, maintaining a 0% collision rate even during aggressive human movements.

Beyond Physical Contact: Psychological Safety

Safety is not merely the absence of collision; it is also the absence of stress. Akansel Cosgun identified the "startle effect" that occurs when silent mobile robots emerge unexpectedly from behind corners. By using Augmented Reality (AR) to project a robot's intended path onto a worker's field of view, researchers can significantly reduce mental stress and improve perceived safety.

To quantify this, Бо Лю and others developed a Psychological Safety Field model. By measuring skin conductance (EDA) to track physiological arousal, they created empirical formulas that allow robots to plan motions that are not just physically safe, but psychologically comfortable. This prevents secondary accidents, such as a worker flinching into a different hazard due to a robot's sudden, albeit non-colliding, movement.

Force Awareness and Compliance

In close-contact tasks like dressing assistance or feeding, physical contact is inevitable. Here, the research shifts toward compliant control and force awareness. Rudolf Lioutikov developed FA-ProDMP, a system that allows robots to distinguish between the resistance of a mechanical obstacle and the soft touch of a human.

This is critical because a robot programmed to "push harder" to overcome environmental resistance could be lethal if it confuses a human arm for a stuck drawer. Similarly, Jihong Zhu and Rajat Kumar Jenamani emphasize that for tasks near the human face, such as feeding (the FEAST system), safety must be verified through rigorous formal methods and LLM-guided code synthesis to ensure that dynamically generated behaviors never exceed safe force limits.

The Frontier: Chaotic Environments

The final frontier for this research is moving from the "sterile" factory to "chaotic" sites like construction. Jingdao Chen and the ICRA Future of Construction Workshop participants argue that construction requires a shift toward human-centric workflows. In these environments, robots like the ALPINE climbing robot or Yuki Sakaida's autonomous excavators must not only avoid humans but also mimic the "hidden" techniques of expert operators to ensure predictable, stable interactions on unstable terrain.

Papers in this version

Showing 60 of 73 papers.

History & citationVersion 1 · 2026-10-04

No longer present in the current build · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_safety_centric_interaction_for_collabora_d5bd4087,
  title = {Safety Centric Interaction for Collaborative Robotics},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/safety-centric-interaction-for-collaborative-robotics/rdr_d5bd4087d78449a6b58eb9b70840feca},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_safety_centric_interaction_for_collabora_d5bd4087_v1,
  title = {Safety Centric Interaction for Collaborative Robotics},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/safety-centric-interaction-for-collaborative-robotics/rdr_d5bd4087d78449a6b58eb9b70840feca/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}