If you are a fan of science fiction, you are probably already familiar with Isaac Asimov’s works, which are set in a world inhabited by humanoid robots. The expansion of business activities by a number of robot manufacturers means that humanoid robots, often known as humanoids, might not remain a figment of our collective imagination for much longer.
Hanson Robotics, the company that created one of the most well-known humanoids, Sophia, has said that the year 2022 will mark the beginning of mass production of thousands of humanoid robots for use in the healthcare industry1. The COVID-19 pandemic has increased the urgency with which such robots are required, particularly in the healthcare, retail, and hospitality industries, in order to lower the danger of infection.
According to the World Robotics 2020 Industrial Robots report2, there are currently a record-breaking 2.7 million industrial robots working in industries all over the world.
Robots are not going away any time soon.
The Importance of AI and ML in the Development of Humanoids
The development of new technologies like artificial intelligence, natural language processing, and machine learning led to considerable improvements in the humanoid robots of successive generations. The following are some areas where artificial intelligence and machine learning play a vital role in the current generation of humanoids:
Vision for Robots
The years 1980 to 1990 were crucial to the development of the idea of robot vision. Robot Vision involves programming a robot to “see” the world in the same way that humans do and interacting with items in two stages:
- Imaging: Use the cameras that are installed in the robot to scan objects in two and three dimensions.
- Image Processing: Once the image has been captured, the robot will next process it using various AI and ML methods. Some examples of these algorithms are dictionary-based item detection algorithms, boosted cascade classifiers, and convolution neural networks.
Pepper, the humanoid created by SoftBank, has a capability called mask detection that can scan up to five faces in a group and determine whether or not the individuals are wearing masks3. Click on this link to obtain the algorithm that is employed by Softbank.
Robots that learn by imitation
In 1999, imitation learning was implemented into the process of training robots, and now, it is a component of reinforced learning. It is comparable to observational learning, in which young children learn how to participate in tasks by watching older people do them. A robot can be taught to carry out a task by the use of demonstrations, which make up the training set, and through the process of learning a mapping between observations and actions through imitation learning. After that, it replicates the behaviors it observes in the presentation in order to accomplish the ultimate objective. Imitation learning has been a significant contributor to the improvement of robots’ motor abilities and the reduction of the amount of human supervision required. The following is a list of some of the fascinating new developments that have come about as a result of applying imitation learning:
Learning how to patch mock human wounds with an accuracy of 85 percent by studying recordings of surgical techniques opens the door for robots to be used in surgery, according to the Da Vinci System4 technology. Surgeons will have more time to concentrate on activities that are more important since robots will do repetitive jobs.
Make it possible for a Robot to Learn a Language5: In a presentation given at a conference in 2018, MIT researchers discussed a parser that could learn language structure by studying captioned movies. The robot makes the connection between the words and the recorded actions and items. If the parser were given a new sentence, it would be able to reliably predict the meaning of the sentence even without the video.
Robots that are capable of Self-Supervised Learning
Robots can enhance their own performance through the use of self-supervised learning, which gives them the ability to construct their own training sets. Robots can learn to scale vast volumes of unlabeled data in a manner that allows for lifelong learning and reduces dataset bias when they use self-supervised learning. Anomaly detection, object recognition, obstacle avoidance, learning new tasks, and many other things are all examples of applications for this technology.
One such example is the Watch Bot, which was built by researchers from Cornell and Stanford6. In order to determine what constitutes “typical human behavior,” it employs a 3D sensor, a camera, a laptop, and a laser pointer. After that, it utilizes a laser pointer to target the object as a reminder (for example, the milk that was left out of the refrigerator), and it has successfully reminded humans of activities that they may have forgotten about sixty percent of the time. The researchers had widened their efforts by giving their robot access to web films in order for it to learn.
Robots that are capable of Multi-Agent Learning
A multi-agent system is a computerized system that is built of numerous intelligent agents that communicate with one another and can solve issues that would be difficult or impossible for a single agent to handle. Robots are given the ability to collaborate and gain knowledge from one another thanks to a concept called multi-agent learning. The use of multi-agent learning in the realm of military research is a particularly fascinating example of its use.
The Centre for Artificial Intelligence and Robotics (CAIR) has been working on developing a Multi-Agent Robotics Framework (MARF)7 in order to provide India’s armed forces with a variety of robots that are capable of cooperating with one another in order to carry out a variety of military tasks including surveillance, rescue, and mapping operations. This will make it possible for a team of different robots that the Indian Army has previously constructed to work together. Some of these robots include a Wheeled Robot with Passive Suspension, a Snake Robot, a Legged Robot, a Wall-Climbing Robot, and a Robot Sentry.
RoboCup8, which features contests such as Robocup Soccer, Robocup Rescue, and Robocup Industrial, is an example of an event that demonstrates the implementation of MARL in robotics. Robots are organized into teams, and those teams battle against one another in an effort to emerge victorious.
The Future, Including Humans and Humanoid Robots
As we move farther into the future, we anticipate the further development of humanoids and robots, as well as the discovery of new applications for these technologies as they become more accessible to a larger percentage of the population. It is reasonable to conclude, given the improvements that have taken place, that humanoids and robots will help automate professions that are repetitious, which runs counter to the popular belief that robots will steal our employment. This will enable humans to focus on activities that are more difficult and lead to the creation of additional jobs. It would be necessary for humans to oversee robots, make use of big data to get deeper insights into the world around them, make predictions about the future, and continue to pursue their individual passions and hobbies.