AI Robot Learns How to Help People Get Dressed

Every day, more than 1 million people in the United States require physical assistance to get dressed, whether because of injury, permanent disability, age…

NVIDIA logo
NVIDIA
•
Nefi Alarcon
•2 min read•beginner•
--
•View Original
KerasTensorFlow

Overview

Researchers from Georgia Tech developed a deep learning-equipped robot named PR2 to assist over 1 million people in the U.S. who require help getting dressed. The robot learns to understand the forces involved in dressing by simulating the experience, using NVIDIA Tesla V100 GPUs and deep learning frameworks like Keras and TensorFlow.

What You'll Learn

1

How to train a robot using deep learning for physical assistance tasks

2

Why simulating human experiences can improve robot assistance capabilities

3

How to utilize NVIDIA Tesla V100 GPUs for deep learning applications

Prerequisites & Requirements

  • Understanding of deep learning concepts
  • Familiarity with Keras and TensorFlow frameworks(optional)

Key Questions Answered

How does the robot learn to assist with dressing?
The robot learns to assist by analyzing nearly 11,000 simulated examples of dressing a human arm. It estimates the forces applied during assistance, allowing it to predict the consequences of different movements and select comfortable motions for dressing.
What technology was used to train the robot?
The robot was trained using NVIDIA Tesla V100 GPUs on the Amazon Web Services cloud, utilizing cuDNN-accelerated Keras and TensorFlow deep learning frameworks to process training data.
What is the current capability of the robot in dressing assistance?
Currently, the robot can put a gown on one arm in about 10 seconds. However, fully dressing a person is a more complex task that the researchers are still working towards.

Key Statistics & Figures

Number of people in the U.S. requiring dressing assistance
1 million
This statistic highlights the significant need for robotic assistance in daily activities.
Number of simulated examples analyzed for training
11,000
The robot's training involved a substantial dataset to learn effective dressing techniques.
Time taken to dress one arm
10 seconds
This demonstrates the robot's current capability in assisting with dressing tasks.

Technologies & Tools

Some links below are affiliate links. We may earn a commission if you make a purchase.

Hardware
Nvidia Tesla V100
Used for training the robot's deep learning models.
Software
Keras
Deep learning framework used for building and training the robot's models.
Software
Tensorflow
Deep learning framework utilized alongside Keras for training the robot.
Cloud
AWS
Cloud platform used for hosting the training environment.

Key Actionable Insights

1
Consider implementing deep learning simulations to enhance robot training for physical assistance tasks.
Simulations provide robots with a way to learn from virtual experiences, mimicking human sensations, which can significantly improve their effectiveness in real-world applications.
2
Utilize high-performance GPUs like the NVIDIA Tesla V100 for training complex AI models.
These GPUs can handle extensive computations required for deep learning, accelerating the training process and enabling more sophisticated models.
3
Focus on understanding the human experience when designing robotic assistance systems.
By incorporating human perspectives into the training process, robots can better anticipate and respond to the needs of users, leading to more effective assistance.

Common Pitfalls

1
Overlooking the importance of simulating human experiences in robot training.
Without considering the human perspective, robots may not effectively anticipate user needs, leading to less effective assistance.