FetchHound
We study how human–dog teams coordinate through gestures, movement and attention, to develop and test models for more natural human–robot collaboration. This page follows our two tasks, pointing and collaborative object search, through our data pipeline: recorded behavior, reconstructed movement and gaze, and model estimates of belief.
How to read a card
Each card is one measure. Every number is a subject’s own average first and then an average across subjects, so n counts animals and children, never trials.
The rule under each value compares the three groups on that card only: full width is the largest of the three, not 1.0, and it is drawn only where the measure is safe to compare across rigs. Click a card for how it is computed and what the numbers say.
The three groups share no subject. Dogs appear in two of them and they are not the same dogs, so every comparison across columns is between subjects, and the owner contrast swaps the pointer and the animals together. The contrast to trust is the within-subject one at the foot of a card: inner cup against outer, on the same individual.
dog · experimenter is the whole pairing, not one task. 24 dogs on the two-camera rig and 6 on the extra-camera PVPE sessions, 238 + 58 trials. Those PVPE sessions are a production study and appear nowhere else on this board, so this column is six dogs and fifty-eight trials wider than the blue pairing on the pointing tab.
“Not recorded” is not zero. Three gaze columns are all-zero on the owner rig while the same columns vary from 0 to 58 elsewhere; showing those as 0 would claim owners’ dogs never look at their person. A cell the rig never populated says not recorded; one the measure does not exist for says no data.
The bar is neither an error bar nor a percentage. ± is the standard error over subjects. Only the 713 verified trials are counted — coded keep and review-approved.
The task
A human stands behind four cups and points at one; the dog, or the toddler, walks up and chooses. Three pairings: an experimenter pointing for a dog, the dog's own owner pointing for it, an experimenter pointing for a toddler.
A human stands behind four cups and points at one; the dog walks up and chooses. Two pairings: an experimenter pointing for a dog, and the dog's own owner pointing for it.
The data
Supplementary: every counted trial of one pairing in one frame, leaving on the first step and arriving on the first cup together.
The task
A dog and its owner look for an object hidden somewhere in a room, both wearing eye trackers, under two conditions: the person leads the search, or the dog does. Every team does both, in counterbalanced order, and a search ends when the object is found or after two minutes.
Objective. Examine how dogs work together with a human to find hidden objects. Dog owners will complete the Monash Dog Owner Relationship Scale (MDORS). The MDORS has been repeatedly developed and validated to measure dog owners’ relationship, with a particular emphasis on dog-owner interaction and perceived closeness (Dwyer et al., 2006). During the task, we will use a dual-eye tracking paradigm, which has previously been used to explore parent-child interactions (Yu & Smith, 2013).
Human-dog teams will work together to find a series of hidden objects under 2 conditions. All human-dog teams will complete both conditions (order of first condition counterbalanced), and for each search teams will have up to 2 minutes. If the hidden object is found prior to that time limit, the trial is considered successful and reset for the next trial. If they’ve completed the Dog IQ survey already for another project in our lab (Ross et al., in prep) that data will be retained. If they have not yet completed that survey, it will be given prior to starting.
The data
A replay of where the dog and the person move and look as they search; person-led and dog-led examples show how each partner's behavior changes with their role.
Mapthe dog (orange) and the person (blue): their trails, where each is looking, and the hiding places in the room
Model estimateseach partner's belief about where the object is hidden, read from their movement and gaze, plus head direction for the dog: the bars for right now, the river for the whole search