💻 computer science

Beyond Collision Avoidance: Multi-Robot Yielding and Spatial Affordance in Emergency Evacuations

Through a virtual evacuation experiment with 56 participants, this study demonstrates that safe multi-robot emergency behavior requires moving beyond pure collision avoidance to semantically aware navigation that proactively utilizes environmental affordances, such as refuge niches, to align with human spatial expectations and minimize perceived cognitive delays.

Original authors: Ning Zhou, Edmund R. Hunt, Nikolai W. F. Bode

Published 2026-05-18

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Ning Zhou, Edmund R. Hunt, Nikolai W. F. Bode

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are rushing out of a building because of a fire alarm. You are running down a narrow hallway, and suddenly, a group of four robots is blocking your path. How they move (or don't move) in that split second could make you feel calm and safe, or panicked and annoyed.

This paper is a study about exactly that scenario. The researchers wanted to figure out the best way for a group of robots to get out of the way of a single person during an emergency, without just crashing into them.

Here is the breakdown of their study, using simple analogies:

The Four "Robot Moves"

The researchers tested four different ways the robots could behave when they saw a human coming:

The "Freeze" (Collective Freeze): The robots just stop dead in their tracks, like statues.
- The Analogy: Imagine a group of people in a hallway suddenly turning into mannequins. You have to squeeze around them.
The "Shortest Path" (Efficiency First): The robots ignore the human and keep running straight down the middle of the hall to get to the exit as fast as possible.
- The Analogy: This is like a group of runners who refuse to step aside for a pedestrian, thinking, "We have a race to win, you move."
The "Line Escape" (Wall Hugging): The robots move to the side of the wall to let the human pass, but they stay in the main hallway.
- The Analogy: Like people in a crowded theater who step to the edge of the aisle but still block the flow a little bit.
The "Hide" (Proactive Yielding): The robots actively look for a small nook, alcove, or recess in the wall and step completely inside it to get out of the way.
- The Analogy: Like a person seeing a small coat closet or a deep doorway and stepping all the way inside so the hallway is completely clear for you.

The Two Hallway Setups

They tested these moves in two types of hallways:

The Empty Hall: A flat, boring corridor with no side spots to hide in.
The Nook Hall: A corridor with small "refuge niches" (little alcoves) built into the walls, perfect for a robot to step into.

What They Found (The Results)

1. The "Hide" Strategy is the Winner
When the robots used the "Hide" strategy (stepping into the nooks), people felt the safest, least anxious, and least obstructed. It was the clear favorite.

Why? It's like a host at a party who doesn't just step aside, but actually goes into the kitchen to let the guest walk through the living room. It feels polite and considerate.

2. The "Expectation Violation" (The Surprising Twist)
This is the most interesting part. When the hallway had nooks, but the robots chose the "Line Escape" strategy (staying in the hallway near the wall), people got more annoyed than if the robots had just frozen or run straight at them.

The Analogy: Imagine you see a clear, empty chair in a crowded room. You expect someone to sit in it. If they stand right next to it instead, you feel a weird sense of frustration, even though they aren't blocking you.
The paper calls this an "Expectation Violation." Because the nook was there, people expected the robot to use it. When the robot didn't, it felt "rude" or "clueless," even though the robot wasn't physically blocking the path. The human brain thought, "It could have hidden, but it didn't!"

3. Experience Matters
People who had interacted with robots before understood the "Hide" strategy even better. They seemed to "get" the robot's social intent faster. However, even people who had never met a robot before still preferred the "Hide" strategy over the others.

The Big Takeaway

The main lesson from this paper is that safety isn't just about physics; it's about psychology.

In an emergency, a robot can't just be a machine that calculates the fastest route (like a GPS). It needs to be a "social" machine.

If a robot sees a little nook in the wall, it should use it.
If it ignores the nook, people will feel more stressed, even if the robot isn't actually hitting them.

The researchers conclude that for robots to be safe and accepted in our lives, they need to understand "spatial affordances." That's a fancy way of saying: "Robots need to know that a little hole in the wall isn't just empty space; it's a place to hide so humans can pass."

They didn't test this with real fire or real crowds, but in a video game simulation. The goal was to understand the basic human reaction before building real robots for real emergencies.

Technical Summary: Beyond Collision Avoidance: Multi-Robot Yielding and Spatial Affordance in Emergency Evacuations

1. Problem Statement

As mobile service robots increasingly coexist with pedestrians, ensuring passively safe behavior during confined emergency evacuations is critical. While existing multi-robot yielding strategies often prioritize macroscopic flow optimization and collision avoidance, they frequently overlook environmental affordances (physical features that suggest specific actions, such as refuge niches) and human spatial expectations.

Current industry defaults often involve a "Collective Freeze" (robots stopping upon detecting crowds), which, while safe for isolated units, physically impedes evacuees in narrow corridors. Conversely, efficiency-first strategies (e.g., robots navigating the shortest path) may optimize flow but violate human social norms. The core problem is the gap between macroscopic trajectory optimization and micro-level human-robot spatial negotiation, specifically regarding how robots should behave when environmental features (like niches) offer clear opportunities for yielding.

2. Methodology

The authors conducted a game-based virtual evacuation experiment ( $N=56$ ) to isolate micro-level human-robot interactions from complex crowd dynamics.

Experimental Design: A $2 \times 4$ $2 \times 4$ mixed-design study.
- Between-subjects factor: Environment Condition. Two architectural layouts: No Affordance (A0) (flat corridor) and Affordance (A1) (corridor with physical refuge niches sized to accommodate the robot).
- Within-subjects factor: Robot Strategy. Four distinct multi-robot yielding behaviors:
  1. Hide: Robots actively retreat into refuge niches (if available) or pull to the wall (if not).
  2. LineEscape: Robots yield to the wall and move along the boundary to the exit.
  3. Freeze: Robots halt at their initial locations (Collective Freeze).
  4. ShortestPath: Robots navigate directly down the center of the corridor to the exit.
Procedure: Participants navigated a virtual hotel corridor (2.2m $\times$ 30m) under fire alarm urgency. They encountered a swarm of four virtual robots executing one of the four strategies.
Data Collection:
- Subjective: 8-item Likert scale questionnaires assessing obstruction, efficiency, cognitive load (NASA-TLX), safety (Godspeed), and social evaluation.
- Objective: Spatial-temporal trajectory metrics (e.g., minimum clearance distance, path straightness, freeze time).
- Moderators: Prior robot experience and niche awareness.

3. Key Results

A. Preference Hierarchy

The study established a robust, statistically significant preference hierarchy for robot yielding strategies in confined spaces:
Hide > LineEscape > Freeze > ShortestPath

Hide significantly outperformed all other strategies in mitigating negative user experiences (perceived obstruction, cognitive load, and agitation).
ShortestPath was consistently ranked last, perceived as the most obstructive, selfish, and anxiety-inducing.

B. The Role of Environmental Affordances

Positive Amplification: When the Hide strategy was used in the presence of niches (A1), it significantly increased perceived calmness and objective physical clearance (minimum distance increased from 0.81m to 1.66m).
Expectation Violation: The LineEscape strategy revealed a critical psychological phenomenon. In the presence of niches (A1), participants rated LineEscape as significantly less efficient and more delaying, despite objective trajectory data showing no physical obstruction or deviation compared to the No Affordance condition.
- The authors interpret this as an Expectation Violation: When a clear niche exists, humans expect a socially competent robot to use it. Failing to utilize an obvious affordance (even if the robot moves efficiently along the wall) triggers a negative psychological penalty.
- In contrast, Freeze and ShortestPath did not suffer this specific penalty for ignoring niches, likely because their behaviors already signal low spatial agency or selfishness, establishing a "low expectation baseline" that cannot be violated further.

C. Moderating Factors

Prior Experience: Participants with prior robot interaction experience rated the Hide strategy as significantly less obstructive and less anxiety-inducing than novices, suggesting experience aids in decoding complex social intents.
Niche Awareness: Conscious awareness of the niches did not significantly moderate the results, supporting the theory that affordance processing operates at a pre-attentive cognitive level.

4. Key Contributions

Comparative Evaluation: Provided a rigorous evaluation of proactive yielding (Hide) versus efficiency-first (ShortestPath, LineEscape) and safety-first (Freeze) strategies in confined emergencies.
Affordance-Expectation Link: Demonstrated that environmental affordances fundamentally shape cognitive expectations. The failure to utilize an obvious niche can trigger Expectation Violation, leading to increased perceived cognitive delay even when physical flow is unimpeded.
Experience as a Decoder: Showed that prior human-robot interaction experience helps users decode complex social intents, though the proactive yielding strategy remains effective even for novices.

5. Significance and Claims

The paper argues that safe human-robot interaction during emergencies must evolve from pure trajectory optimization to semantically aware navigation.

Passive Safety Redefined: Passive safety is not merely about avoiding collisions or optimizing flow; it requires respecting spatial semantics. In high-stakes scenarios, a robot's failure to utilize available refuge spaces generates psychological friction and cognitive load, undermining trust.
Design Implication: Future multi-robot social navigation algorithms should be semantic-aware, dynamically mapping environmental affordances to switch between macro-level flow optimization and micro-level proactive yielding based on crowd density and spatial context.
Limitations: The authors acknowledge that the study focused on micro-level, single-person interactions and virtual environments. They note that crowd dynamics in high-density scenarios and physical embodiment factors (noise, collision risk) may alter these findings, necessitating future real-world validation.

The research concludes that for socially acceptable coexistence, robots must prioritize yielding space over trajectory efficiency in confined, high-stress environments, aligning their behavior with human spatial intuition and implicit social obligations.