This article discusses features and benefits of robots that promise to change the way therapists help children and adults with autism and pave the way for more social robots in the future. Robots can enable therapists to deliver treatment more effectively, and can prompt classroom teachers, aides, and parents with limited training through the steps needed to teach an autistic child a skill. This could make intensive therapy more accessible and affordable. The Milo robot works as a force multiplier, reinforcing skills taught by human teachers. PABI’s electronic and servo insides are guarded by spring steel ribs and a plush, stuffed exterior. Autism robots can manage mechanics automatically, and they typically link with a tablet and as the child chooses the cards, the robot automatically records data more accurately than any human could do and immediately displays new cards. While Milo and PABI are very different robots, designers want to build robots that adapt to individual children. They hope to do this by adding better sensors and algorithms.


Laurie Dickstein-Fischer discovered how robots might help autistic children at a robotics conference in Japan in 2009. She was pursuing a doctorate in school psychology, and attended the meeting with her husband, Gregory Fischer, a mechanical engineer creating surgical robots at Worcester Polytechnic Institute in Massachusetts.

In Japan, both were drawn to the displays of social robots, which used sensors and artificial intelligence to perceive and interact with people. Most were costly and, in Dickstein-Fischer’s opinion, served no real purpose.

The exception was Paro, a furry seal-shaped robot that responded to its name and purred and vibrated when petted. Adults with dementia loved it and treated it like a pet. But unlike a living pet, it needed no care or feeding and never misbehaved.

At the meeting, Dickstein-Fischer met an American engineer who started talking about using robots for autism intervention. “He talked about autism as a pathology,” she said. “I had worked with autistic students in Baltimore, and thought of them as people, not pathologies.

“But I came home thinking this was the next big thing, using robots to work with autistic children. I said to Greg, “You’re an engineer, I’m a psychologist. We could build a robot especially for autistic children from the ground up.’”

She envisioned a robot that looked like a child’s stuffed animal, a penguin with large eyes and exaggerated features. Back at WPI, she and Fischer put together a group of students to build PABI, the Penguin for Autism Behavioral Intervention.


Around the same time, Carolyn Garver, who had directed the Autism Treatment Center in Dallas for 30 years, got a call from David Hanson, a well-known roboticist who had worked for Disney Imagineering.

“He said that he had a robot and thought it would have implications for autism,” Garver recalled. She never heard of Hanson and knew nothing about robots, but invited him over. Hanson brought Milo, a robotic face to her office and cabled it to his laptop.

The face contained motors and flexible skin. It smiled, frowned, looked confused or angry, and moved like an animatronic character in a Disney theme park.

“The minute I saw it, I loved it,” she said. Hanson and his chief engineer, Richard Margolin, teamed with Garver to test how people with autism would respond. The tests confirmed Garver’s enthusiasm. “Children and adults were really taken by them,” she said. “One person even asked the robot on a date.”

Milo and PABI are as different as two robots can be. Margolin believes Milo’s expressive face engages people, while Dickstein-Fischer thinks it might scare some people. Yet both robots can help teach autistic children and adults practical and social skills.

Margolin eventually bought Hanson’s autism operation and renamed it RoboKind. He has begun selling Milo robots for autistic children. Meanwhile, PABI has moved through several design iterations, and Dickstein-Fischer is planning additional upgrades and tests.

Other autism robots are just reaching the market or are in development. Some, like Darwin-OP2 by Chung Hyuk Park, an assistant biomedical engineering professor at George Washington University, are humanoids. Others are less conventional, like Paris-based Leka’s ball-mounted display, which uses AI to play games with autistic and developmental^ challenged children. Several developers have adapted other social robots, such as Blue Frog’s Buddy and SoftBank Robotics’ Nao, for autism intervention.

The torrent of new autism robots is not surprising. Over the past 10 years, better processors, sensors, and AI algorithms have made social robots far more responsive. Autism therapy gives them a purpose—and offers a structure that suits their emerging skills. In many ways, autism therapy is a first step on the way to true social robots.

The Spectrum

No one is certain how many people have autism. Recent estimates by the Centers for Disease Control and the more in-depth National Health Interview Survey estimate it is 1.5 to 2 percent of all children from three to 17 year old.

Autism is difficult to pin down because it expresses itself in many ways. Most autistic people struggle with social skills, from knowing when to shake hands to identifying when others are happy or sad. They find it difficult to communicate, control their emotions, and apply what they have learned to new situations. They often rebel at change.

The intensity of symptoms varies so widely, experts now call this condition autism spectrum disorder. More than one-third of people on the spectrum have severe autism, and are intellectually challenged or non-verbal. Others, such as actress Daryl Hannah and comedian Dan Ackroyd, have symptoms that are barely noticeable.

Most people with autism fall somewhere in the middle, and many of them have jobs, including accountants, laboratory technicians, and software programmers. A substantial percentage succeeded because interventions early in life helped them manage their behaviors. A child who learns how to decode the emotions on a face or respond appropriately in social situations will find it easier to fit into school and eventually the world.

Unfortunately, early intervention takes trained therapists and time. This makes it expensive. Most schools and parents cannot afford daily therapy.


Robots can help. They enable therapists to deliver treatment more effectively, and can prompt classroom teachers, aides, and parents with limited training through the steps needed to teach an autistic child a skill. This could make intensive therapy more accessible and affordable.

In fact, it was the potential of robots to reduce the cost of early interventions that initially drew Dickstein-Fischer to the field.

“When I was a teacher and school counselor in inner city Baltimore, where incomes averaged $17,000, I did not have the resources to serve all those children and neither did their parents,” she said. “I thought that a $2,000 robot that schools could afford would make it possible to deliver those services.”

Studies by Dickstein-Fischer and others show that robots can do this, and how they do it goes to the heart of the human-robot relationship.

Robotic Reinforcement

Autistic children struggle to pick up social clues that they might help them make sense of the world. As a result, therapists must teach them things that other children pick up naturally. This could be as simple as identifying an apple or when someone looks happy or sad.

The most common way to teach that skill involves placing three cards on the table. They might show an apple, an orange, and a pear. The therapist or teacher asks the child to point to the apple. If the child picks right, the teacher says, “Great job.” If not, the teacher prompts the child to try again, trying to remain positive.

It sounds easy. But while the teacher is doing this, he or she is also measuring the child’s progress by timing responses with a stopwatch, recording answers and prompts, and gathering and reshuffling the cards. All of this pulls the teacher’s attention away from the child. Therapy mechanics, Dickstein-Fischer said, are a major cause of teacher and therapist burnout.

Autism robots can manage these mechanics automatically. They typically link with a tablet. As the child chooses the cards, the robot automatically records data more accurately than any human and immediately displays new cards.

“When we tested the robot, not only did the kids become more engaged, but we got an unexpected result—it cut the time needed to complete the therapy in half,” DicksteinFischer said.


This is because the robot recorded the therapy data faster than the teacher could. It also freed the teacher to focus exclusively on the child.

“Pairing PABI with the tablet worked better than the tablet alone, and students and teachers were more on task,” she said. “They were visibly happier, and there was more interaction between child and therapist.”

The robots also prompted the children to stay focused.

“We have a video of a girl hugging PABI and rubbing its tummy,” Dickstein-Fischer said. “We never saw that child interacting with a stuffed animal before. When she gets an answer right, the teacher praises her and PABI says, 'Great job,’ and flaps its wings up and down. When it’s wrong, it asks her to try again. For her, it’s like getting reinforcement from a friend.”


Autism robots provoke those feelings because they have an embodied physical presence and they are responsive. They act as if they were alive. While most realize they are not really alive, studies by educators, roboticists, and psychologists show that people treat them very differently than computers or tablets. This is why people with dementia fight over who gets to hold the Paro seal, and why children want to hug PABI.

Autism robots build on that engagement. “They are very consistent, and do the same thing over and over without ever getting upset or having a bad day,” Garver said. “For autistic children and adults, a robot is predictable and easier to deal with than other people.”

Robots can also go beyond teaching facts to helping children deal with their feelings, recognize emotions in others, and navigate new social situations.

For example, Garver and Pamela Rollins, an associate professor at the University of Texas’ Callier Center for Communication Disorders in Dallas, developed a curriculum to help autistic children calm down. This included teaching children how to identify when they were getting upset and a series of skills, like taking deep breaths, counting to 10, or squeezing a ball, to manage those feelings.

They also taught children to recognize the emotions of people around them.

“Autistic kids don’t understand the perspectives of others,” Garver said. “So, we showed Milo being happy, then asked them to pick out the happy person on the tablet and Milo would say, “Right,” if they got it right. The display on Milo’s chest might show someone smiling or even the child smiling. And all the time, there is a facilitator sitting there, encouraging and motivating them.”

They used these skills on Milo to help children navigate social situations.

“Kids on spectrum will say, ‘hello,’ but not look at you or smile,” Rollins explained. “Now, you and I know that before you talk to someone, you need their mutual attention, so you look at them, smile, and say ‘hi.’

So, Rollins developed a series of social narratives, short stories and images Milo can show on its chest display, that describe a social situation, clues about how to identify the situation, and appropriate ways to respond.

“We teach this the way someone taught us math by describing skills like how to add or subtract,” she said. “Icons on the robot’s chest say and show that when you greet someone, you look at them, smile, and say, 'hi.’

“So, we describe the situation, then we show different children and adults looking, smiling, and saying hi on the tablet. Milo asks the children, 'Did my friends do it right?’ The child with autism must discriminate if they did all the components.

“Milo never gets angry, it’s always positive. If a child does something wrong, it says, 'Let’s do it again.’ And it seems to be working. About 30 percent of kids who were not making progress by other means are making progress with the robot,” Rollins said.


Building Milo and PABI presented several challenges. First, they had to be rugged enough to stand up to kids.

Fischer encased PABI in spring steel ribs that pop back into shape when squeezed, surrounded by a soft, compliant stuffed animal body. He combined 3-D printed mechanical components with off-the-shelf hobbytype servos to pan and tilt the head and individual eyes. The wings use antagonistic pairs of cables to flex up and down or point, but, as Fischer noted, that’s enough to give a high-five.

While Fischer does a lot of work in his lab, his home has become a museum of penguin prototypes. There’s a 3-D printer and soldering station in the basement, offset by piles of fluffy fabric and opened pillows used for stuffing.

A colleague from the university who sews costumes for theater stitches everything together, and Fischer and Dickstein-Fischer’s two-year-old son helps with product testing.

Margolin spent years making Milo more robust and affordable in a lab that has its own combination of order and chaos. Along the way, he brought in a consultant who specialized in mechanical toys to recommend gearboxes, motors, and simplified control systems.

“It’s not a pick-and-place robot, so we didn’t need that kind of accuracy,” Margolin said.


He also found other ways to shave costs. Milo uses stiff rather than bendable legs that give it a “Frankenstein” walk but simplify design and reduce problems with overheated motors and gearbox jams. But he also added spring-loaded safety clutches that keep the robot from exerting more than 2.5 lb of force.

Margolin spent most of his time on Milo’s expressive face. “In terms of engagement, kids really like the face,” he said. “We’re actively teaching social and emotional skills, and autistic kids connect to it better than they do to an animated screen.”

Margolin uses six servos, most of which move through arc, plus one linear motor driving a rack-and-pinion system to control facial movement. He molds the face with a custom-formulated polymer that curves and twists, aided by embedded hard points that tweak those motions to get the expressions exactly right.

While Milo and PABI are very different robots, Margolin and Fischer both want to build robots that adapt to individual children. They hope to do this by adding better sensors and algorithms.

Both are working on video cameras that track a child’s eyes and bring them back on task when they wander. Both hope to use cameras and microphones to tell if a child is happy or sad, so they can respond appropriately.

“We are also looking at very low powered radar to monitor heart rate, breathing, and gestures,” Margolin said. “When fused with data from video and voice, it will enable the robot to create a theory of mind about the child’s emotional state and respond to that.”

This is just the beginning, Fischer said. “One of the reasons I like the type of child training we do is that we almost want teachers to be robotic,” he said. “It very structured, but maybe we can use this to learn what we need to get into a less structured environment.

“Maybe one day, instead of flashcards on a tablet, we’ll have a bucket of toys that we spill out on the floor and the robot says, 'Touch the red dinosaur.’”