💻 computer science

Warmth and Competence in the Swarm: Designing Effective Human-Robot Teams

This paper demonstrates that in human-robot swarm teams, specific behavioral manipulations like broadcast duration and separation distance significantly influence human perceptions of warmth and competence, which ultimately drive team preferences more strongly than actual task performance.

Original authors: Genki Miyauchi, Roderich Groß, Chaona Chen

Published 2026-04-22

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Genki Miyauchi, Roderich Groß, Chaona Chen

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 watching a school of fish swim in a tank, or a flock of birds turning in unison in the sky. You don't just see them as machines or animals; you instinctively judge their "personality." Are they acting like a helpful, friendly group? Or are they acting like a cold, efficient, but distant machine?

This paper asks a very modern question: If a group of robots works together (a "swarm"), do humans judge them the same way we judge people? And more importantly, does how we feel about them matter more than how well they actually do their job?

Here is the breakdown of the research, explained with some everyday analogies.

The Big Idea: The "Warmth vs. Competence" Scale

Psychologists have long known that when we meet someone (or a group), we judge them on two main things:

Warmth: Are they friendly? Do they have good intentions? (Like a helpful neighbor).
Competence: Are they smart? Can they get the job done? (Like a skilled surgeon).

Usually, we want both. But if we have to choose, we often prefer someone who is warm and competent over someone who is just a cold, efficient robot.

The Experiment: The Robot "Search Party"

The researchers created a simulation where 10 little robots had to find a hidden "treasure" (a target zone) in a square room. They had to work together to find it and get there as fast as possible.

To see how humans perceived them, the researchers tweaked three "knobs" on the robots:

Speed: How fast they moved.
Spacing: How far apart they stayed from each other.
Broadcasting: How long they waited to tell the others where the treasure was once they found it.

They ran two different tests:

Study 1 (The Audience): People sat back and watched the robots on a screen.
Study 2 (The Teammate): People actually controlled one of the robots, trying to help the group find the treasure.

The Surprising Findings

1. The "Helpful Neighbor" Effect (Warmth)

The Finding: The longer a robot waited to share the treasure's location with the others, the warming the group seemed.
The Analogy: Imagine you find a great restaurant.

If you immediately run to the kitchen to tell the chef, you seem efficient but maybe a bit frantic.
If you stop, look around, and make sure your friends know about it before you go, you seem caring and cooperative.
The robots that "broadcasted" (shared info) for a longer time looked like they were prioritizing the group over themselves. Humans saw this as "warmth."

2. The "Spread Out" Effect (Competence)

The Finding: The robots that kept a larger distance from each other were seen as more competent.
The Analogy: Think of a search party looking for a lost hiker in a forest.

If the searchers are all huddled in a tight circle, it looks like they are confused or stuck.
If they spread out wide to cover more ground, it looks like a strategic, organized, and capable team.
Humans interpreted the wide spacing as a sign of intelligence and good planning.

3. The Speed Trap

The Finding: Surprisingly, how fast the robots moved didn't change how people felt about them.
The Analogy: A race car driver might be fast, but if they are rude, you don't trust them. Conversely, a slow driver who is polite is still liked. The speed of the robots was irrelevant to their "personality."

The Twist: Watching vs. Doing

When people were just watching (Study 1), they judged the robots based on the rules above.
When people were part of the team (Study 2), the feelings shifted slightly.

The "Humble" Shift: When people joined a "High Warmth/High Competence" team, they actually rated the team lower than the observers did. It's like when you join a perfect team and realize, "Wow, we have a lot of work to do," making the team feel less magical.
The "Underdog" Boost: When people joined a "Low Warmth/Low Competence" team, they rated them higher. It's like the "we're all in this together" feeling; when you struggle with a team, you start to appreciate them more.

The Most Important Lesson: Feelings > Facts

The biggest takeaway is this: People preferred robot teams that felt warm and competent, even if those teams were slower at the task.

The Analogy: Imagine you need to hire a moving company.

Team A is incredibly fast and efficient but rude and ignores your questions.
Team B is slightly slower but smiles, explains what they are doing, and helps you carry the boxes.
Most people would hire Team B.

The study found that humans are the same with robots. If a robot team is fast but acts cold, humans won't want to work with them. If they act friendly and smart (even if they take a few extra seconds), humans will trust and prefer them.

Why This Matters

As we start using swarms of robots for things like search-and-rescue, farming, or delivery, we can't just program them to be the fastest. We have to program them to be socially aware.

To look smart: Make them spread out and cover ground.
To look friendly: Make them pause to share information with the group.

The paper concludes that for humans to trust and work well with robot swarms, we need to design them not just as machines, but as good teammates.

1. Problem Statement

As robot swarms increasingly collaborate with humans, the design of these systems has traditionally focused on optimizing objective performance metrics (e.g., task completion time, efficiency). However, there is a significant gap in understanding how humans socially perceive robot swarms. While prior research has established that humans evaluate individual agents based on warmth (intent, friendliness, helpfulness) and competence (ability, skill, intelligence), it remains unclear:

Whether these social dimensions apply to collective swarm behaviors.
How specific swarm parameters (speed, separation, communication) map to these perceptions.
Whether these perceptions differ when a human acts as a passive observer versus an active team member (operator).
How social perception influences team preference compared to objective task performance.

2. Methodology

Experimental Setup & Simulation

Platform: The study utilized the ARGoS simulator to model a homogeneous swarm of 10 e-puck robots (7 cm diameter, differential drive) in a 150 cm × 150 cm arena.
Task: A collective search task where robots must explore the environment, discover a hidden circular target region (radius 25 cm), and all converge on it as quickly as possible.
Behavioral Model: Robots operated in three modes:
1. Explore: Ballistic motion with obstacle avoidance.
2. Share Target: Upon finding the target, a robot broadcasts the location to neighbors for a duration $T$ .
3. Move to Target: Robots flock toward the target using virtual potential forces.
SwarmUI: A custom web-based interface was developed to allow human participants to control one robot within the swarm (Study 2) while the rest operated autonomously.

Experimental Design

The researchers conducted two studies manipulating three core swarm parameters:

Speed ( $v$ ): 5.0 to 15.0 cm/s.
Separation Distance ( $d$ ): 4 to 36 cm (distance robots try to maintain from neighbors).
Broadcast Duration ( $T$ ): 0 to 16 seconds (how long a robot shares target info).

These parameters generated 125 distinct team configurations.

Study 1 (Observer): 90 online participants watched video clips of 15 randomly selected swarm teams. They rated each team on Warmth, Competence, and Team Preference.
Study 2 (Operator): 16 participants performed the task in a lab, controlling one robot via SwarmUI. They interacted with teams selected from four quadrants (High/Low Warmth $\times$ High/Low Competence) derived from Study 1 data. They rated Warmth, Competence, and Joint Effort.

3. Key Contributions

Social Perception Framework for Swarms: The paper establishes that warmth and competence are valid, measurable dimensions for evaluating robot swarms, not just individual agents.
Parameter-to-Perception Mapping: It identifies specific technical parameters that reliably influence social perception:
- Broadcast Duration $\rightarrow$ Warmth.
- Separation Distance $\rightarrow$ Competence.
- Speed $\rightarrow$ No significant effect on either dimension.
Role Invariance vs. Modulation: The study demonstrates that while the overall pattern of perception (high vs. low warmth/competence) remains consistent whether the human is an observer or an operator, active involvement (being a team member) slightly modulates the intensity of these ratings.
Preference Drivers: It provides empirical evidence that social perception is a stronger predictor of human team preference than objective task performance.
Open-Source Tool: The release of SwarmUI to facilitate future human-swarm interaction research.

4. Key Results

Influence of Parameters on Perception (Linear Mixed-Effects Models)

Warmth: Significantly predicted by Broadcast Duration ( $p < 0.001$ ). Longer durations (robots "helping" each other by sharing info longer) were perceived as warmer. Robot speed and separation distance had no significant effect.
Competence: Significantly predicted by Separation Distance ( $p = 0.01$ ). Larger separation distances (more dispersed formations) were perceived as more competent, likely suggesting better exploration capabilities. Broadcast duration and speed had no significant effect.
Speed: Had no significant effect on either warmth or competence ratings.

Impact on Team Preference

Social Perception > Performance: When predicting team preference, Warmth ( $\beta = 0.74$ ) and Competence ( $\beta = 0.92$ ) were strong predictors. Task Performance (completion time) was a significant but much weaker predictor ( $\beta = 0.12$ ).
Conclusion: Humans prefer teams that are perceived as socially warm and competent, even if they are not the fastest.

Observer vs. Operator Dynamics

Consistency: The relative ranking of teams (e.g., High Warmth/High Competence vs. Low/Low) remained robust across both studies.
Shift in Ratings: Active participation shifted absolute ratings.
- High Warmth/High Competence teams were rated lower in Study 2 (operator) than in Study 1 (observer).
- Low Warmth/Low Competence teams were rated higher in Study 2.
- This suggests that direct involvement introduces a "halo effect" or increased empathy, slightly elevating the perception of struggling teams, while perhaps making high-performing teams feel less "special" due to the operator's own contribution.

5. Significance and Implications

Design Guidelines: Swarm designers should prioritize information sharing duration to convey warmth and spatial distribution to convey competence, rather than focusing solely on speed.
Human-Centric Collaboration: The findings challenge the "efficiency-first" paradigm in robotics. To build trust and effective collaboration, systems must be designed to be socially engaging. A robot swarm that is fast but perceived as "cold" or "incompetent" may be rejected by human partners.
Dynamic Perception: The study highlights that social perception is dynamic; it changes based on the human's role (observer vs. participant), necessitating adaptive interfaces or behaviors depending on the level of human involvement.
Future Research: The work opens avenues for designing swarms that balance technical efficacy with social acceptability, using the identified parameters as levers for tuning social behavior.