Distributed multi-robot systems for the home often require robots to operate out of the user's sight, creating a state awareness gap that can diminish trust and perceived transparency and control. This paper investigates whether real-time, socially mediated state externalization can bridge this gap without compromising task performance. We developed a system where a co-located social mediator robot (Pepper) externalizes the hidden execution states of an out-of-sight mobile manipulator (Stretch~3) for voice-driven object retrieval and delivery, where task-level states are synchronized and externalized through verbal updates and visual progress display. In a counterbalanced within-subject study (N=30), we compared a baseline of Autonomous Hidden Execution against Socially Mediated State Externalization. Our results show that externalization significantly increases user task-focused attention (from 15.8% to 84.6%, p<.001) and substantially improves perceived perspicuity, dependability, stimulation, and attractiveness (all p<.001). Furthermore, 83% of participants preferred the externalized condition, and this improvement in user experience was achieved without a statistically significant increase in end-to-end task completion time (p=.271). The results suggest that socially mediated state externalization is an effective architectural mechanism for designing more transparent and trustworthy distributed robot systems, ultimately enhancing user experience without sacrificing performance in distributed home robot deployments.
Figures & tables
Fig. 1: Overview of the distributed execution architecture and state externalization framework. The user interacts with a co-located social mediator, while task execution is performed by a mobile manipulation robot operating in a physically separated workspace. A coordination server synchronizes task-level execution states (e.g., NAVIGATING, GRASPING, RECOVERING) and externalizes them to the user through verbal and visual feedback. This architecture enables real-time state awareness despite spatial separation.
Fig. 2: System architecture of the distributed robot interaction framework. User voice commands are captured by the Pepper social mediator, which forwards structured intents to a coordination server. The server dispatches task commands to the Stretch mobile manipulator for navigation and object retrieval. During execution, the robot publishes task-level state updates (e.g., NAVIGATING, SEARCHING, GRASPING), which are externalized to the user through Pepper via verbal and visual feedback. When a failure is detected, a closed-loop recovery process is triggered: the failure event is externalized to the user, user confirmation is collected, and a recovery command is issued to the execution robot.
State
Description
IDLE
The robot is awaiting task assignment.
NAVIGATING
The robot is moving toward a target location.
SEARCHING
The robot is locating the requested object.
GRASPING
The robot is attempting object manipulation.
FAILED
A navigation or manipulation failure has been detected.
RECOVERING
The robot is executing a recovery strategy.
TABLE I: Task-level execution state definitions.
Fig. 3: Experimental layout of the distributed task execution setup under two conditions. (a) Condition A : Autonomous Hidden Execution: the execution robot operates in a physically separated workspace without communicating intermediate task states to the user. (b) Condition B : Socially Mediated State Externalization: a co-located social mediator provides real-time verbal and visual updates about the execution robot’s task states. In both conditions, the execution robot performs object retrieval in a separate room, while the user remains spatially separated from task execution. The purple arrow indicates the execution robot’s navigation path during retrieval and delivery tasks.
Fig. 4: Example end-to-end trial timeline for Condition B (Socially Mediated State Externalization). (a) The participant issues a request via the social mediator (Pepper) and confirms the task. (b) While the execution robot operates in a separate workspace, the mediator externalizes task progress through verbal updates and a visual progress display. (c) The execution robot retrieves the target object and returns. (d) The item is delivered to the participant, and the mediator concludes the interaction.
Metric
Description
Task Initiation Time
Duration from system readiness for speech input to dispatch of a structured task command.
Execution Time
Time from task dispatch to completion (or failure), including navigation and manipulation phases.
End-to-End Task Time
Total duration from user request initiation to final task completion or termination.
Number of Grasp Attempts
Total grasp trials executed during a task session.
Task Success
Binary indicator of successful object retrieval and delivery.
Task-Focused Attention Ratio
Proportion of execution time during which participants oriented toward task-relevant areas, coded from video.
TABLE II: Objective Performance Metrics
Dimension
Description
Perspicuity
Understanding of system behavior and awareness of ongoing task states.
Dependability
Perceived reliability and sense of control during execution.
Stimulation
Engagement and feeling accompanied during task progress.
Fig. 5: Objective performance and interaction metrics under Hidden Execution and Socially Mediated State Externalization. (a) Time-based metrics, including task initiation time, execution time, and total end-to-end duration. (b) Average number of grasp attempts per task session. (c) Task-focused attention ratio during execution. Error bars represent ± 1 SEM. Asterisks denote statistical significance ( ∗p<0.05 , ∗∗∗p<0.001 ). .
Metric
A Mean (SD)
B Mean (SD)
t(29)
p
Init. (s)
33.47 (7.84)
49.93 (9.12)
9.21
<0.001∗∗∗
Exec. (s)
162.63 (64.54)
137.33 (35.87)
-2.56
0.016∗
Total (s)
196.10 (67.78)
187.26 (38.40)
-1.12
0.271
Grasp Att.
2.20 (1.21)
1.78 (0.93)
-1.54
0.134
Attention (%)
15.8 (8.3)
84.6 (6.5)
21.47
<0.001∗∗∗
Suc. Rate (%)
86.7
90.0
–
0.754
TABLE IV: Objective performance under Autonomous Hidden Execution (A) and Socially Mediated State Externalization (B).
Fig. 6: Object-level task distribution and manipulation success rates. (a) Percentage of object requests across categories (Water, Chips, Fruit). (b) Object-wise task success rates aggregated across conditions. Although success rates varied across object types, object requests were moderately distributed and no condition-specific interaction was observed, indicating that object composition did not confound the transparency effect.
Metric
A Mean (SD)
B Mean (SD)
t(29)
p
Perspicuity
3.12 (0.74)
4.31 (0.62)
6.84
<0.001∗∗∗
Dependability
3.36 (0.69)
4.18 (0.58)
4.97
<0.001∗∗∗
Stimulation
3.08 (0.77)
4.45 (0.51)
8.12
<0.001∗∗∗
Attractiveness
3.24 (0.72)
4.36 (0.55)
6.21
<0.001∗∗∗
TABLE V: Subjective ratings for Autonomous Hidden Execution (A) vs Socially Mediated State Externalization (B).
Outcome
Effect
β
Test stat.
p
Total task time
Condition
7.98
z=0.70
0.484
Period
-2.85
–
0.803
Grasp attempts
Condition
0.44
z=1.89
0.059
Period
-0.26
–
0.274
TABLE VI: Linear mixed-effects analysis of order effects. Fixed effects include Condition (Hidden vs. External) and Period (first vs. second exposure); participant is a random intercept.
Fig. 7: Failure mode breakdown for failed trials under both experimental conditions. Failure events were categorized into navigation errors, grasp failures, system hangs, Pepper-related errors, and other/unknown causes. Although the total number of failed trials differed slightly, the relative composition of failure types remained comparable across conditions, indicating no systematic shift in underlying failure mechanisms.
Assistive robots operating under shared autonomy must balance user control with autonomous assistance. Because robot actions depend on internal intent inference that is not directly observable, mismatches between inferred and intended goals can undermine coordination and trust. We investigate how interface-level transparency, including feedback modality (visual vs. auditory) and information richness (sparse vs. rich), shapes interaction in a vision-based shared autonomy system. In a user study with N=25 participants across two assistive manipulation tasks, we evaluate how these designs influence coordination and trust. Providing feedback significantly improves intent alignment and reduces corrective intervention, indicating that making the inferred goal legible accelerates convergence in shared control. Participants preferred visual over auditory feedback, while preferences for sparse versus rich information depended on task complexity. We also found that revealing the full belief distribution did not consistently improve alignment or trust. Together, these findings indicate that effective transparency enhances coordination primarily through goal legibility, while trust depends on task-appropriate information exposure rather than maximal disclosure. Based on these results, we outline guidelines for designing transparent shared autonomy systems.
Atharv Belsare, Zohre Karimi, Connor Mattson +2
Kahlert School of Computing, University of Utah, Salt Lake City, UT, USA. · Robotics Center, University of Utah, Salt Lake City, UT, USA.
Domestic robots rely on rich perception to operate in private homes, but privacy risk persists even when raw sensor data remain local. Structured representations exported to downstream planners, cloud services, logs, or learning pipelines can still reveal household information through semantics, geometry, spatial structure, and task targets. We introduce Task-Functional Perception Distillation (TFPD), a task-scoped representation-export framework that keeps rich perception local and profiles downstream exports according to task utility, direct exposure, and multiple residual inference risks. Using 120 AI2-THOR scenes with scene-disjoint train/validation/test splits, frozen attacker selection, and representation-aware held-out attacks, we evaluate navigation, collision checking, and object-goal execution. Three navigation exports achieve identical success (1.000) and mean path ratio (0.898), yet representation-level linkability ranges from 0.532 to 0.970. Replacing an explicit target label with a target region reduces target-category macro-F1 from 1.000 to 0.077 while preserving success at 0.995, while geometric coarsening reduces object-category macro-F1 from 0.704 to 0.556 at a measurable collision-utility cost. A ProcTHOR replication preserves the navigation task-equivalence/privacy-inequivalence finding while changing the relative ordering of normalized and topological exports. These results show that neither field removal nor stronger abstraction induces a universal privacy ordering and motivate task-specific, multi-risk evaluation of the complete public representation.
Yuqiao Xu, Erman Ayday
Department of Computer and Data Sciences Case Western Reserve University Cleveland, OH, USA
Robots that assist people must respond promptly and appropriately to human activity, which requires low-latency processing, accurate perception, and anticipation of human motion. These requirements are especially difficult to meet in living environments, where perception must cover wide areas, remain robust to occlusion, and capture natural behavior without wearable devices. We present OmniRobotHome, a testbed that addresses these challenges in a fully furnished home that supports natural everyday activity. Its 48 hardware-synchronized cameras provide room-wide coverage, enabling accurate markerless perception under occlusion. A distributed perception pipeline processes these streams in parallel and delivers persistent human and object states to robot controllers with low latency. We demonstrate OmniRobotHome through two human-robot interaction studies: proactive yielding for safety and nonverbal intent recognition for assistance. Through these studies, we analyze how camera coverage, state latency, multi-view pose, hand, and gaze estimation, and behavior-memory-based forecasting affect overall task performance. Our results highlight the value of OmniRobotHome as a testbed for studying human-robot assistance in real living environments.