Experiments with human subjects
To evaluate how well our method behaves around humans, we also conduct additional experiments with ten human subjects. We recruited ten subjects from among graduate students, visiting scholars, and staffs on campus. We consider the balance to be as diverse in gender (5 male, 5 female), professional (6 student, 4 non-student), and origin (4 North America, 4 Asia, 1 Europe, 1 Latin America) as possible.
We asked the subjects to walk through the space during a robotic navigation trial. Subjects were not told which method was being used for each trial. After each experiment, subjects were asked qualitative questions with a 5-point Likert scale: (1) was the robot motion intrusive? (2) was the robot friendly and natural? (3) did the robot motion feel smooth? (4) did the robot pre-emptively move out of your way?
Here, we show the answer for four questions and some example videos when having experiments with human subjects.
Fig. Â Evaluation of socialness by human rating. The height of each bar indicates the mean and the range line indicates the standard deviation. Larger is better in all ratings. * indicates statistical significance of our method on the t-test.
Our method
Residual RL
Female A
SACSoN with finetune
Our method
Residual RL
Female B
SACSoN with finetune
Our method
Residual RL
Male A
SACSoN with finetune
Our method
Residual RL
Female C
SACSoN with finetune
Our method
Residual RL
Male B
SACSoN with finetune
Our method
Residual RL
Female D
SACSoN with finetune
Our method
Residual RL
Female E