The advent of artificial intelligence (AI) technology is ushering in a new era of robots designed to streamline everyday activities. These AI-powered robots can revolutionize personal and professional lives, offering increased efficiency, improved accessibility, and a transformed way of interacting with technology. This article deliberates on the growing role of AI-powered robots in everyday activities.
Image Credit: dee karen/Shutterstock
Introduction to AI-powered Robots
Applications of AI and machine learning (ML) are undergoing tremendous growth, with endless use cases in real-world scenarios. Robotics, an interdisciplinary domain encompassing computer science and other applied mechanisms, is also gaining significant attention for various applications. It involves the construction, design, and use of ML and AI techniques to perform specific tasks typically performed by humans.
AI-powered robots can assist humans in performing dull, repetitive, and dangerous tasks with less error. They can also make decisions required to execute a task. The major scope of AI in robots designed and developed for everyday activities includes motion control, grasping, and vision. Vision enables robots to detect and recognize previously unseen items.
AI and ML assist robots in determining the best position to grasp an item, and also with motion control to enable them to interact in a dynamic environment and avoid obstacles in their way. Additionally, AI/ML assists robots in understanding data patterns to allow them to act efficiently.
Increasing Importance of AI-powered Robots
AI-powered robots are becoming more prevalent in multiple application areas of real life. They have become more adaptable to new environments and show significant improvements in their performance by incorporating AI techniques. For instance, robotic packaging utilizes AI techniques for more accurate, lower-cost, and quicker packaging.
Similarly, AI-powered chatbots are now being used frequently for customer service in hotels and retail stores. These chatbots harness the AI's natural language processing abilities to interact with customers and answer their queries. In medical science, AI-powered robots are commonly used for the precise execution of tasks. Many clinical examinations and surgeries are currently performed using these robots.
In the field of mental health, chatbots and virtual embodied AI-supported psychotherapeutic devices have been investigated for dealing with depression and anxiety. The industrial sectors utilize AI-powered robots for accurate and efficient management of raw materials, and for cleaning, cutting, polishing, fixing, and precise drilling of different non-metallic and metallic materials.
Such precise execution of routine activities ensures the timely completion of various tasks and error-free productivity. By implementing unmanned aerial vehicle (UAV) control and combining AI and image processing, autonomous inspections have been performed.
AI-powered robots are also employed for transporting patients, disinfecting rooms, handling routine logistical tasks, moving heavy machinery, surveillance, aerial photography, distribution and logistics, product deliveries, policing, increasing personalized data collection, providing new customer experiences, improving pre-arrival experiences, increasing productivity in service organizations, and generating insights.
Generative AI-powered Robots
The use of foundation models is a key idea behind generative AI. These types of foundation models can be referred to as Large X Models (LXMs), where "X" represents the training data category needed to provide the AI with general-purpose capabilities in a particular space.
For instance, "X" can represent text data, machine data, and human action data for language models, machine models, and action models, respectively. The opportunities for LLMs, including potential physical models assisting with household chores and existing chat models, are endless.
For example, a large behavior model trained using data obtained from videos of humans performing physical tasks could be employed to teach robots about the execution of different everyday activities and chores, like peeling potatoes or watering plants.
Studies have shown that diffusion, a generative AI technique utilized for popular text-to-image applications, can be used to teach domestic tasks like peeling vegetables to robots. Generative AI substantially reduces the time required to teach robots compared to traditional programming.
LXM-powered robots can drive productivity gains through machine automation in major domains of physical work, including healthcare, hospitality, domestic work, maintenance, logistics, construction, manufacturing, and agriculture. Recently, humanoid, general-purpose robots have been introduced by Tesla, Sanctuary, Prosper, Figure, Boston Dynamics, and Agility.
For instance, Prosper is building a robotic helper, Alfie, for the office or home. Alfie can organize and clean different items and perform small chores like watering plants. In the future, multiple LXM-powered general-purpose robots are anticipated to be built for diverse tasks. Researchers from Carnegie Mellon University have already enabled robots to learn household chores by watching videos of humans performing daily tasks in their homes.
Recent Developments
A paper recently published in Robotics presented Neural Signal Operated Intelligent Robots (NOIR), an intelligent, general-purpose brain-robot interface (BRI) system that enables humans to command robots through brain signals to perform everyday activities.
Humans can communicate their objects of interest and intended actions to the robots using non-invasive electroencephalography (EEG) through this interface. The principle of this proposed system is hierarchical shared autonomy, where high-level goals are defined by humans and the robot actualizes the goals through low-level motor command execution.
This system distinguishes itself by taking advantage of advancements in ML, robotics, and neuroscience. For instance, NOIR is general-purpose in its diversity of accessibility and tasks, as researchers showed that humans could accomplish 20 daily everyday activities, which is in contrast to existing BRI systems that typically exist only in simulation or specialize in one or few tasks.
The system could also be utilized by the general population with a minimal amount of training. Additionally, the "I" in NOIR implies that the robots are adaptive and intelligent as they are equipped with a collection of diverse skills, enabling them to perform low-level actions without extensive human supervision.
Human behavioral goals could be executed, interpreted, and communicated naturally by robots with parameterized primitive skills. Moreover, the robots can learn human intended goals during collaboration with humans. Thus, the objective of the research was to demonstrate that a more adaptive system with limited data can be developed by leveraging the recent advancements in foundation models. This approach could substantially enhance the system's efficiency.
NOIR can be greatly beneficial for those individuals who require assistance with everyday activities. In this work, researchers selected tasks from the Activities of Daily Living and BEHAVIOR benchmark to effectively capture the actual human requirements.
These tasks included four mobile manipulation tasks and 16 tabletop tasks, encompassing multiple categories, including three entertainment tasks, three personal care tasks, six cleaning tasks, and eight meal preparation tasks. Formal definitions of these activities were provided in the BDDL language format, which specifies the goal and initial conditions of a task using first-order logic, for systematic evaluation of task success.
The proposed system demonstrated success in 20 challenging, everyday household activities, including personal care, cleaning, entertainment, and cooking. Specifically, the synergistic integration of the system with robot learning algorithms improved its effectiveness, enabling the NOIR to adapt to individual users and predict their intentions.
Although the tasks were challenging and long-horizon, NOIR displayed encouraging results by completing tasks with only 1.83 attempts on average. Additionally, the robot learning algorithm improved NOIR's efficiency. Using a simple image classification model using ResNet, an average accuracy of 0.31 was achieved.
However, the proposed method with a pre-trained ResNet backbone realized a substantially higher accuracy of 0.73. This highlights the importance of retrieval-based learning and contrastive learning techniques in improving accuracy. The accuracy further improved to 0.94 when R3M was used as the feature extractor. Researchers also evaluated the generalization ability of the algorithm.
In all variations, the proposed model achieved 93% accuracy, which implied that the human could skip the object and skill selection process 93% of the time, significantly reducing effort and time. Thus, this study enhanced the way of interaction between humans and robots by replacing traditional interaction channels with direct, neural communication.
Overall, AI-powered robots are revolutionizing everyday activities, with brain-controlled robots like NOIR offering a glimpse into the future of intuitive human-robot collaboration. However, challenges like safety and ethical concerns need to be addressed for widespread adoption.
References and Further Reading
Zhang, R. et al. (2023). NOIR: Neural Signal Operated Intelligent Robots for Everyday Activities. ArXiv. DOI: 10.48550/arXiv.2311.01454, https://arxiv.org/abs/2311.01454
Das, S., Das, I., Shaw, R. N., Ghosh, A. (2021). Advance machine learning and artificial intelligence applications in service robot. Artificial Intelligence for Future Generation Robotics, 83-91. DOI: 10.1016/B978-0-323-85498-6.00002-2, https://www.sciencedirect.com/science/article/abs/pii/B9780323854986000022
Parker, G. G., Gupta, A. (2024). What’s next for generative AI: Household chores and more [Online] Available at https://mitsloan.mit.edu/ideas-made-to-matter/whats-next-generative-ai-household-chores-and-more (Accessed on 01 July 2024)
Gonzalez-Aguirre, J. A. et al. (2021). Service Robots: Trends and Technology. Applied Sciences, 11(22), 10702. DOI: 10.3390/app112210702, https://www.mdpi.com/2076-3417/11/22/10702
Ness, S., Shepherd, N. J., Xuan, T. R. (2023). Synergy between AI and robotics: A comprehensive integration. Asian Journal of Research in Computer Science, 16(4), 80-94. DOI:10.9734/ajrcos/2023/v16i4372, https://journalajrcos.com/index.php/AJRCOS/article/view/372