When looking at the list of video presentations, the one on “Building Caregiving Robots” by Tapomayukh Bhattacharjee, fondly known as Tapo, caught my attention the most because I was a caregiver to my grandmother a few years ago. So, it was interesting to learn how a robot would go about caring for people with mobility limitations. As an Assistant Professor at Cornell University, Tapo leads the Enterprise Lab, where his research revolves around enabling robots to assist individuals with mobility limitations in their daily activities. His work spans the vast terrain of human-robot interaction, haptic perception, and robotic manipulation, all with the noble aim of leveraging technology for the betterment of society.
The Challenge of Caregiving
Tapo’s presentation shed light on the daunting challenge of caregiving in today’s world. There are approximately 24 million adults in the United States alone who require assistance with various daily tasks, ranging from feeding to dressing and meal preparation. However, although it looks simple, caregiving is an intricate and context-dependent problem. Care recipients exhibit diverse functioning abilities and behavioral traits.
For instance, in one scenario, a caregiver might need to place food close to the recipient’s mouth, while in another, it must be precisely positioned above their left molar. Moreover, care recipients may exhibit varying levels of engagement, with some engrossed in TV screens while others look to their caregivers for cues. Transferring care recipients between positions and environments adds another layer of complexity.
Why Caregiving Robots?
The question naturally arises: do we really need robots to perform these tasks? Several compelling reasons advocate for their use. Firstly, robots can offer a level of control and independence to care recipients that may be otherwise unattainable. For example, while a human caregiver can help the care recipient, they might have limitations like not being available all the time. However, a robot is always available to the patient. Additionally, while human caregivers play a crucial role, care recipients might still feel uncomfortable, rushed, or socially challenged. Moreover, caregiving is time-consuming and can contribute to caregiver burden, making it a multifaceted and technically demanding problem.
It’s important to note that building caregiving robots is no small feat. Apart from the intricate nature of the tasks, several barriers hinder progress, including limited access to care recipients, realistic environments, and the high cost of robot hardware.
Simulating Progress
To overcome these hurdles, Tapo and his team turned to simulations. They developed “RCareWorld” and collaborated with institutions like Columbia University to gather crucial data, such as the range of motion of patients. They trained the robots using simulation environments and then transferred the acquired skills to the real world.
One notable initiative is “CareHomes,” which provides home environment assessments modified for assistive purposes, along with the necessary assistive devices. These modifications span three levels of accessibility:
- Level 1: “Usual,” with no special modifications.
- Level 2: “Partially barrier-free,” involving adjustments to frequently used objects.
- Level 3: “Completely barrier-free,” where all possible modifications are implemented.
RCare also delves into muscular skeletal modeling and interfaces for attaching force sensors to robot arms, enabling comprehensive sensing capabilities. They experimented with reinforcement learning in simulation for tasks like dressing, toileting, limb repositioning, and feeding. Notably, they successfully transferred a bed bathing policy trained on a human avatar with soft tissues to the real world.
The Role of Simulation
But is simulation the answer? While it may not replace real-world testing, simulation serves as a valuable tool for prototyping and problem-solving. It allows researchers to iterate and refine their approaches before implementing them in the field.
Introducing SPARCS
To bridge the gap between laboratory research and real-world needs, Tapo introduced “SPARCS: Structuring Physically Assistive Robotics for Caregiving with Stakeholders-in-the-loop.” This approach emphasizes understanding the human side of caregiving first, recognizing its contextual nature. It considers the user, the caregiver, the environment, and the robot as interconnected elements.
The Complexity of Robot Caregiving
Robot caregiving is multifaceted and deeply contextual, relying on the user’s needs, their environment, and the caregiver’s role. SPARCS takes a structured approach:
- Understanding human caregiving: Building blocks and task overflow.
- Understanding robot caregiving: Instantiating building blocks and task workflow.
- Action workflow: Translating human caregiving to robot caregiving.
Robot-Assisted Feeding: A Complex Challenge
One specific area of focus is robot-assisted feeding, a particularly challenging endeavor. It involves various phases, including bite acquisition, bite timing, and bite transfer. The technology’s robustness is critical, as it must handle different textures and deformable objects. In other words, the robot must be able to lift food that it has not seen before.
Tapo’s team trained a neural network on 20 food items. However, this is problematic as there are 4000+ food options in the real world. So, to make the model robust, the team must train the robot on a wider range of food dataset.
Training a robot neural network on few food items makes the robot prone to errors. What I found interesting is that users with high mobility limitations preferred more autonomy, even with occasional errors while those with fewer limitations had higher expectations and accepted partial autonomy. For your information, there are three levels of autonomy: full autonomy, low autonomy, partial autonomy. Full autonomy means user selects food items through an interface and all phases are automated. Low autonomy means user selects food item through an interface but the robot requests input in every phase. Lastly, partial autonomy means user selects food item through an interface and only one phase is automated.
At the end of the lecture, Tapo discussed how to be a roboticist. He explained a lot of the subjects that students learn in high school are important such as mathematics (probability, statistics in the basis of A.I., matrix computations, calculus, etc.), physics (statics, dynamics, etc.), and so on. Additionally, skills that every wanna-be roboticist should practice are analytical skills, coding skills (ROS, Python, C++).
My key takeaways from this assignment are that robotics is a high-impact field as there are many ways in which robots can help humans/humanity. However, since robotics is a new field, challenges exist. Without watching this presentation, I would not have come up with the challenges that faced the roboticist who created the assistive feeding robots.
In conclusion, my appreciation for the field of robotics has increased after this presentation, and I’m now aware of the challenges that might arise when training a robot in even simples tasks like handling daily tasks such as feeding.
Author: Ali Ramazani
Computer Science Major @ Berea College