Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2019 Nov 6:6:110.
doi: 10.3389/frobt.2019.00110. eCollection 2019.

Long-Term Personalization of an In-Home Socially Assistive Robot for Children With Autism Spectrum Disorders

Caitlyn Clabaugh  1 , Kartik Mahajan  1 , Shomik Jain  1 , Roxanna Pakkar  1 , David Becerra  1 , Zhonghao Shi  1 , Eric Deng  1 , Rhianna Lee  1 , Gisele Ragusa  2 , Maja Matarić  1
Affiliations

Long-Term Personalization of an In-Home Socially Assistive Robot for Children With Autism Spectrum Disorders

Caitlyn Clabaugh et al. Front Robot AI. .

Abstract

Socially assistive robots (SAR) have shown great potential to augment the social and educational development of children with autism spectrum disorders (ASD). As SAR continues to substantiate itself as an effective enhancement to human intervention, researchers have sought to study its longitudinal impacts in real-world environments, including the home. Computational personalization stands out as a central computational challenge as it is necessary to enable SAR systems to adapt to each child's unique and changing needs. Toward that end, we formalized personalization as a hierarchical human robot learning framework (hHRL) consisting of five controllers (disclosure, promise, instruction, feedback, and inquiry) mediated by a meta-controller that utilized reinforcement learning to personalize instruction challenge levels and robot feedback based on each user's unique learning patterns. We instantiated and evaluated the approach in a study with 17 children with ASD, aged 3-7 years old, over month-long interventions in their homes. Our findings demonstrate that the fully autonomous SAR system was able to personalize its instruction and feedback over time to each child's proficiency. As a result, every child participant showed improvements in targeted skills and long-term retention of intervention content. Moreover, all child users were engaged for a majority of the intervention, and their families reported the SAR system to be useful and adaptable. In summary, our results show that autonomous, personalized SAR interventions are both feasible and effective in providing long-term in-home developmental support for children with diverse learning needs.

Keywords: autism spectrum disorders; early childhood; home robot; long-term human-robot interaction; personalization; reinforcement learning; socially assistive robotics.

PubMed Disclaimer

Figures

Figure 1
Figure 1
The hierarchical framework for human-robot learning (hHRL) comprises of a two-level hierarchy: (1) The meta-controller takes high-level information about the current state of the intervention and activates a lower-level controller. (2) The lower-level controllers await activation to select the robot's action based on a simplified state representation, reward, and action category: instructions, promises, feedback, disclosures, and inquiries.
Figure 2
Figure 2
The real-world long-term SAR intervention for early childhood math learning used a meta-controller that sequentially executed each controller. At the beginning of a session, the robot disclosed that it needed the child's help to reach a specific planet. It then promised that they would reach the planet if they completed all the games they needed to do that day. The child and robot played 10 games whose challenge (pLoC) and feedback (pLoF) levels were set by the computational personalization methods. At the end of each session, the robot congratulated the child on completing the games and reaching the promised planet. It then asked the child some open-ended questions about their day.
Figure 3
Figure 3
The physical in-home setup included the SPRITE robot with the Kiwi skin (C) mounted on top of the container encasing a computer, power supply, and speakers (E), with an easy-access power switch (D), a camera (B), and touchscreen monitor (A), all located on a standard child-sized table (F).
Figure 4
Figure 4
The child-robot interaction was designed around Kiwi as a robot space explorer. The following diagram displays varying challenge levels of the Pack Moon-Rocks game, with more challenging problems combining math reasoning and numerical operation concepts.
Figure 5
Figure 5
The reinforcement learning reward for the personalized level of challenge (pLoC) ranged between −5 and 5, where 5 indicated that the user was completing games at the highest challenge level. The average cumulative reward of pLoC matched each participant's pre-intervention scores of the WIAT II subtests for numerical operations and math reasoning. Therefore, the pLoC adapted to each participant over the month-long intervention (there is no pLoC data for P1 or P2, as explained in section 4.2).
Figure 6
Figure 6
The reinforcement learning reward for personalized level of feedback (pLoF) ranged between 0 and 1, where 1 indicated that a child completed games with the least amount of support or feedback. The average cumulative reward of pLoF converged over 25–50 episodes (i.e., mistakes and help requests) and is correlated with the intervention length and the average number of mistakes made by child participants per game (there is no pLoF data for P1 or P2, as explained in section 4.2).
Figure 7
Figure 7
Participant survey results for adaptability (left) and usefulness (right).
Figure 8
Figure 8
Overall engagement for each analyzed participant, with error bars denoting standard deviation across sessions (left). High variance but no linear trend (p = 0.99) in engagement is observed across sessions (right) (engagement was only analyzed for participants with adequate video and audio data).
Figure 9
Figure 9
Variance in engagement is higher across participants than across LoC (left). Participants with high optimal LoC were more engaged (rs = 0.84, p = 0.018) (right). Engagement was only analyzed for participants with adequate video and audio data.
Figure 10
Figure 10
All participants (excluding P3 who did not complete the study) showed significant improvements (d = 0.54, p < 0.01) on the WIAT II subtests for numerical operations (left), and math reasoning (right) between the pre-intervention and post-intervention assessments.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Adams E. (2013). Fundamentals of Game Design. 3rd Edn. Pearson, GA: New Riders.
    1. Amabile T. M. (1993). Motivational synergy: toward new conceptualizations of intrinsic and extrinsic motivation in the workplace. Hum. Resour. Manage. Rev. 3, 185–201.
    1. Amershi S., Cakmak M., Knox W. B., Kulesza T. (2014). Power to the people: the role of humans in interactive machine learning. AI Mag. 35, 105–120. 10.1609/aimag.v35i4.2513 - DOI
    1. Anderson J. R. (1985). Cognitive Psychology and Its Implications. New York, NY: WH Freeman; Times Books; Henry Holt & Co.
    1. Anderson J. R., Reder L. M., Simon H. A. (1996). Situated learning and education. Educ. Res. 25, 5–11.