Sci-Fi on the Factory Floor: Brain-machine Interfaces

With the advances in artificial intelligence (AI) and machine learning (ML), there has also been a rise in brain-machine interface (BMI) research. Just recently, researchers at the University of Technology Sydney (UTS) designed a non-invasive brain-machine interface (BMI) technology. The technology allowed users to operate a quadruped robot with their minds, showing 94% accuracy. 

While robotic "mind control" is still a technology in research and development, a fully-integrated BMI would result in having a thought acted out in real life just by merely thinking of it. BMIs have the potential to improve reaction times and perform appropriate actions, perhaps without requiring as much job-specific training.

 

Researchers at UTS used non-invasive sensors to enable robotic “mind control.” Image used courtesy of ACS Applied Nano Materials

 

What Is a Brain-machine Interface (BMI)?

A BMI uses some sort of biosensor, either embedded in the human being or in a harness or helmet, to collect feedback from brain signals. Depending on which areas of the brain are activated, the sensors will receive these signals and process them into robotic action elsewhere. The BMI consists of several parts: the sensor, the processor, and the robotics to perform the action. All three of these areas need to develop together to create a full BMI, and all three have unique challenges before a full BMI can be implemented.

 

Challenges in Sensors

Sensing technology for BMI applications is difficult. First, sensors will either need to be surgically implanted in the person or be sensitive enough to measure the brain’s signals through the skull while also filtering out noise signals in our increasingly noisy electromagnetic environment. Surgical implants will not move around as much as a helmet or harness, though the procedure is quite invasive and looks more like the plot from a dystopian novel.

Furthermore, the sensors will either need a long battery life, such as is found in pacemakers, or will need to be a passive sensor. However, even a passive sensor will need to transmit data through the skull to the processor. Harnesses and helmets are not as invasive, but they will likely need to be calibrated for each user to ensure that the sensors are placed over the same spot on the head every time. They will be misleadingly easy to transfer between people, but the actual implementation will be tricky.

 

UTS researchers designed a headband with non-invasive sensors. Image used courtesy of ACS Applied Nano Materials

 

Challenges in Processing

Regardless of the current state of affairs in computing power, the human brain is still much more powerful. The limitations in processing power are due to our ability to harness it. This is why there is so much research into teaching styles, learning styles, and other such topics, as these can expand the learning capacity.

Despite the big talk of AI, AI is still bound through its programming. It is not truly a “learning” computer. When all the fancy parts are removed, AI is still a large case statement that responds in an “IF THEN” cause-and-effect relationship. The human brain is a true learning machine. It tries one path and one cause and then checks the effect. If the results are undesirable, it determines when and how to try again, dreaming up new ideas.

A BMI would bypass some of this processing requirement by allowing the human brain to handle the complex learning and some of the decision-making. Motion control could be deferred to the processor. However, the processor would need to handle how to provide feedback to the user and ensure that motions are within certain safeguards.

For example, consider the brief, fleeting thought that crosses the mind when walking behind someone moving too slowly. Instead of punching them in the back of the head automatically, the human brain can determine that this would be a poor choice and quickly filter out that action. Some of these actions would need to be filtered by the external processor instead.

 

Challenges in Motion Control

The robotic motion itself faces the fewest challenges, as numerous automation solutions can be activated through control signals from a processor. However, certain safeguards in motion control require a feedback signal.

For humans, the simple act of picking up an object and moving it requires feedback from pressure in the fingertips to determine how tightly to apply the force of the hand when grasping the object. Robots can perform this action as well, as the feedback mechanism feeds directly into the processor.

Consider a robot that senses that it should not move in a certain direction for fear of collision. The processor will not allow it to move into an occupied space. If an operator attempts to move the robot into this space, the processor says, “no,” and the robot does not move any farther. The operator checks and sees why but then issues new commands. With a BMI, while thinking “the robot should move here,” the operator may continue to send the “move here” signal while trying to figure out why the robot is not moving. It may actually delay the response time.

 

BMIs could potentially increase response time and decrease the time spent in training for specific processes. Image used courtesy of Adobe Stock

 

BMI Benefits and Potential Uses

BMIs have the potential to reduce the response time for certain actions. Currently, a human’s response time consists of seeing an event, mentally processing it, and sending commands to the fingers and hands to interact with a human-machine interface, all while walking through a Standard Operating Procedure, which can change with time and differ from the initial training procedure. With a BMI, a plant operator can see an event, process it, and then allow the external processor to decide which actions need to be performed and in what order based on the newest code version.

The real benefit of BMIs is adding some abilities to people with disabilities. Spinal cord injuries and defects prevent stimuli from reaching the brain and control signals from reaching the rest of the body. With a BMI, there may be a way to bypass an injury.

Consider an accident victim who has had a leg amputated and a broken spinal cord. Not only are they missing a leg, but a prosthetic will not help, as they have no means to control the muscle to move the prosthetic. With a tuned BMI, their brain, which is still trying to send control signals to the missing leg, can be used to manipulate a prosthetic. Potentially, this process could occur with only minimal retraining and physical therapy.

 

Final Thoughts

One might ask, “Are BMIs a good thing for humanity?” Overall, the answer is probably yes. They can potentially increase response time and decrease the time spent in training for specific processes. They could bring mobility to those who have never had it or lost it through disease or accident.

While BMIs are still very early in development, these advantages will take much more time to be fully realized. During this development time, it will be imperative for companies developing these systems to apply good practices and engineering ethics to prevent BMIs from being misused or as a method to invade privacy. Furthermore, for safety purposes, there needs to be adequate action filtering to prevent every single thought from becoming a potentially regrettable action.