Добавил:
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:

Readings in Robotics Engineering. Учебное пособие

.pdf
Скачиваний:
0
Добавлен:
06.09.2026
Размер:
2 Мб
Скачать
2.5. Using biomimicry to build robust robots Video
61
2.5. Using biomimicry to build robust robots Video
Natural organisms, through millions of years of natural selection, have evolved to take on the challenges of their environment. These attributes make them adaptable, versatile and robust and it is these design goals that researchers aim to achieve in robotics.
In this video, Dr Sean Anderson explains how the answer to a biological question - why can’t we tickle ourselves? - helped researchers at The Universi­ty of Sheffield to solve a particular robotic problem.
Transcript
0:05 One of the problems we have in robotics research is what happens when we take the robots out of the laboratory environment and into the real world? And at that point, we have a multitude of sensory information that could potentially confuse the robot. One example of this is how does the robot distin­guish between signals that it causes itself and signals that are caused by interac­tion of the robot with the external world? A nice example of this in humans is tickling. Why can’t we tickle ourselves?
0:33 And the issue there is, basically, it’s thought by the neuroscientists that we predict the consequences of our own motor actions and the sensory consequences of those actions, so if you like, the touch that we feel from our
own fingers moving. But we can’t predict how the sensations will be caused by
someone else tickling us. That’s an unpredictable thing. And so it’s thought that if we put this kind of algorithm into a robot, the robot will be much better able to distinguish between signals that it causes itself, or sensations that it causes itself, and sensations that are caused by the external world. A nice example of this is in our whisking robot.
1:13 So, we set up an experiment where we had our whisking robot use its whiskers to tactilely investigate the world. The idea was that it was supposed to contact an object and then orient, or turn, towards that object. What we found was that worked fine, except on some occasions it would turn and orient to something that was not there, and we called this a phantom orient, or a ghost orient. The way we corrected this was to set up an algorithm in the robot where it can actually predict the sensory consequences of its own actions. So it could predict the sensory signals generated by its own movement of its whiskers.
Part 2. Learning from nature
62
1:48 And by doing that, it could then distinguish which sensations were caused by its own whisker movements, and which sensations were caused by contact with the external world, because those contacts with the external world were unpredictable.
Using mind control
Recent progress in understanding Brain-Computer Interface (BCI) tech­nology is opening the doors for new robotic applications. We look at some of the research that is taking place at Sheffield to empower people with disabilities.
2.6. Brain controlled robots
An exciting, emerging area where biology meets robotics is in brain computer interfaces (BCIs). BCIs have the potential in the future to enable di­rect mind control of robots and other machines. This article explores the defini­tion of a BCI and explains the key components.
Brain computer interfaces
In the past few years, BCIs have attracted a lot of attention from robot­ic groups, neuroscientists, computer scientists and neurologists, triggered by new scientific progress in understanding brain functions and by impressive applications.
BCI Definition: A BCI is a device that measures electrical activity in the brain or spinal cord, and transforms it into a synthetic output that can re­place, restore or enhance natural function.
Based on this definition, a BCI system can control a robot or other assis­tive devices using our thoughts. Such a system can greatly help people with generalised paralysis to gain some level of independence.
Brain computer interfaces for robot control
The BCI input is the brain signals that carry informative neural fea­tures, recorded by electrodes either in the brain or on the head. The BCI out­puts are used to control a device, such as an assistive robot, a wheelchair or a prosthetic hand.
Example: A person suffering from paralysis from the neck down would normally find independent mobility impossible. With a BCI system and a ro­botic wheelchair, the person can use mental imagery to imagine movement of
2.6. Brain controlled robots
63
their right hand in order to turn a wheelchair towards their right side, or their left hand to turn the chair to their left.
The BCI uses algorithms to translate the measured brain-wave activity into command signals to control the output device, i.e. in this case, the motors driving the wheelchair to the left or right.
One of the key challenges of BCIs to solve in the future is decoding brain-wave activity into desired ac­tions. This problem is usually ad­dressed with machine learning algo-
rithms, known as ‘classification’ (also known as ‘pattern recognition’). This is
where a particular set of brain-wave
patterns are ‘classed’ as a specific ac-
tion, e.g. move left or move right.
Key components of a BCI
The whole architecture of an online BCI system is summarised in the di-
agram below:
The core components of a BCI system are as follows:
1. Measurement of brain activity This part is responsible for recording brain activities using various types of sensors. After amplification and digitisa­tion, the recorded brain signals serve as BCI inputs.
2. Preprocessing This unit reduces noise and artifacts present in the brain signals in order to enhance the relevant information hidden in the input signals.
3. Feature extraction The feature extractor transforms the preprocessed signals into feature values that correspond to the underlying neurological mecha­nism. These features are employed by BCI for controlling the output device.
4. Classification This part is responsible for identifying the intention of the user from the extracted features.
5. Control a device The output device can be a computer, a wheelchair or a robotic arm etc. The output of the classifier is used as a command to con­trol the output device.
Part 2. Learning from nature
64
6. Feedback The BCI should feedback the consequences of the action to the user, in a closed loop, so that the user can make adjustments. Feedback can be in visual, auditory or tactile form.
All these units are highly important in the development of an efficient BCI and affect the BCI performance in terms of accuracy, speed, and infor­mation transfer rate. A BCI must be designed to comfortably carry out this pro­cess without any harm to the user’s health.
Reading: ROBOT BRAINS
To complete their tasks, robots have to sense, plan and act. Robots use different kinds of sensors to collect the information they need. Software pro­cesses this information so the robot can plan a response. Then they act to get the job done.
Robots need to have someone tell them what to do. This process is re­ferred to as programming: the way that we can make robots or computers fol­low instructions. Any action a robot is going to do needs to be specifically pro-
grammed for it to complete the task. A robot “thinks” and learns by processing
data and then uses this information to plan its actions.
Programming is the source of instructions for the robot. A robot’s pro-
gram is a set of instructions that tells it what to do, how to do it and when to do it. In order for the robot to complete a task it must be programmed to do so.
Programming requires defining a robot’s task as a series of logically based
step-by-step instructions that can be followed sequentially in order to reach a goal. A robot’s program will also contain a library of simple commands that allow a programmer to describe things in the same way every time. Robots cannot interpret variability in commands the way humans do. For example, a
robot will not be able to differentiate between a command to “sit down” versus “please sit.
Emphasizing the specificity of the command is important. At times, the
robot’s environment may be strewn with obstacles or unexpected events. Pro-
grammers need to think through the possibilities of these various scenarios and plan accordingly so they can communicate in any situation. The tasks a robot completes must be defined up front. For example, if the robot does not know
how to respond to a command to “sit”, then giving the robot that command will
result in no action from the robot.
2.7. Using a brain-computer interface Video
65
A robot will have a library of available actions. Without that action be­ing in its library, the robot is not capable of understanding. One way to think about this is to imagine a programming language that does not have the addi­tion (+) operator that is used in math. In this example, the programmer can nev­er give the computer any math operation that includes addition. Similarly, if the
robot does not have a library of actions that says “lift feet,” then a programming step that includes “lift feet” will result in the robot doing nothing.
Robot software is the collection of coded commands that tell the robot what tasks to perform. Robot software is used to determine what tasks should be performed and to carry out that action. Programming robots can be an intri­cate task. Many software systems and frameworks have been proposed to make programming robots easier.
Programming involves computational thinking: a way to analyze and solve problems. Computational thinking requires deconstructing the entire deci­sion-making process, the variables involved and all possible solutions, ensuring the right decision is made based on the corresponding parameters and limita­tions of the problem. Computational thinking can be useful in almost any situa­tion that requires solving a challenge.
Discuss:
How do people complete a new task they have never done?
How does the robot know what to do or what types of actions to take?
In what ways are robots not smart?
What is holding them back?
https://www.fi.edu/sites/default/files/EducatorGuide_RobotRevolution_Ex hibit_edguide-robot-revolution.pdf
2.7. Using a brain-computer interface Video
Dr Mahnaz Arvaneh is developing a brain-computer interface (BCI) which uses electroencephalography (EEG) technology to read brain signals and convert them into actions on a computer screen.
In this video, Mahnaz demonstrates her system, using the Emotiv EPOC+, an EEG headset which measures brain signals.
Part 2. Learning from nature
66
Discussion
What challenges do you think we could face when working with mind-
controlled systems like this? Why might it be difficult?
Transcript
0:05 The system that we used is called brain-computer interface. So we have brain in one part and computer, or electronic device, in the other part, and the system connects the two together. We record brain signals while our users are doing different types of mental tasks. For example, imagination of move­ment of right hand versus imagination of movement of left hand. We apply a number of signal processing algorithms to improve signal-to-noise ratio. There­after, the relevant patterns of the brain will be extracted. And we use a model, a classifier, to distinguish between different types of mental tasks that have been performed. So the model can distinguish - can identify - which type of mental task has been performed.
0:57 The output of this model can be used for controlling a device. And the good thing is that the user can see what is the output of the device. The user can identify if his performed mental task has been identified, has been recog­nised, correctly or wrongly. So, in other words, the user can also start changing the strategy of doing different mental tasks to improve the performance of the system. The system that we used today is called P300-based BCI, and generally it’s a software that we can use for communication, for spelling words.
1:37 It will be very helpful, especially for people who are locked-in­those who are severely disabled - so they can use this device to communicate with the other people, with the external world. We call it P300 because it works based on a wave in our brain, which is called P300. When we see a target stim­ulus, our brain reacts, our brain generates a peak, which happens usually around 300 milliseconds after the onset of the stimulus. So in this system, as you saw, different letters are flashing. When the desired letter is flashed, the peak hap­pens in the brain of the user. So the algorithm, the system, can capture this peak and translate it to the letter that was expected to be spelled.
2:31 Generally in brain-computer interface, one of the most challenging issues is noise. Our brain signals are very sensitive to body movements, blink­ing, any facial movements, because our muscles are also generating electrical activities. And these electrical activities can interfere our brain signals. When
2.8. What are the possibilities of mind control?
67
our participant blinked, we had a huge artefact in the signal, so we need to de­velop signal processing algorithms that can separate these movement-based artefacts, to focus on the brain signals. Our brains are different. My brain is different than your brain. So the patterns that my brain generates would be dif­ferent than the patterns that the other person generates. It means that the model that we train will be different from person to person.
3:37 We need to collect, usually, a few minutes’ data. And based on that data, we can train our model - we can adjust the parameters - and thereafter, the user gets able to use the model and control a device. Unfortunately irrelevant but concurrent neural activities can affect our brain signals. For example, when we start getting tired our brain patterns also change, so the model that we trained previously might not be optimal anymore. So mood can affect our brain signals, tiredness can affect… It means that, for example, come tomorrow, the model is not optimal anymore. So, we need to, again, adjust the parameters.
4:24 We need to develop, in the future, robust algorithms that can deal with all these variations that we have from time to time, from day to day. Using BCI we can help severely paralysed people to gain some sort of independence. They can use this technology to control their environments. They can have a type of assistive robot that is controlled by their brain signals. Just imagine
those who are paralysed from below the neck, so they don’t have any muscle
movements, but they can use their brain signals to control a robotic hand, for example - that is one of the areas that I’m working in.
5:04 It’s neuroprosthetic - how we can generate different patterns of our hand - grasping, pinching, taking different objects - based on the activities that our brain generates.
2.8. What are the possibilities of mind control?
Available BCI applications can be divided into two main categories. The first and most important category is the medical domain. Indeed, the main ob­jective of BCI is to serve disabled people with a new communication and con­trol channel. The second category is the non-medical domain, for instance by designing video games based on BCI technology.
Some of the applications possible with current BCIs are described below.
Part 2. Learning from nature
68
Personal communication
This group of applications comprises the spelling of text, composing and sending emails and letters. If a patient has generalised paralysis, their only way of communication without a BCI (if any) is with the help of a caregiver or note­taker. Introducing BCI-based communication devices enables the patient to send confidential messages that provides a great amount of independence and self-determination.
Environmental control and control of assistive robots
Opening the front door to visitors, turning lights on or off, controlling shades, regulating the room temperature, changing the backrest position as well as controlling an assistive robot are examples for this group of applications. Their use increases independence and decreases the workload of caregivers.
Rehabilitation training
Recently, motor imagery-based BCI has been proposed as a rehabilita­tion tool to facilitate motor recovery in stroke. Motor imagery (MI) is a mental process in which a subject imagines that they are performing a movement. Sev­eral studies have demonstrated that motor imagery has a positive effect on mo­tor rehabilitation after a stroke through activation of the affected sensorimotor networks. Since the performance of MI is internal to the subject, and thus not directly observable, BCI can facilitate the MI-based stroke rehabilitation by providing direct and immediate feedback on the MI performance.
Control of paralysed limbs
Using a BCI system can lead to the control of a limb orthosis (brace) or even allow for direct control of body parts via functional electrical stimula­tion (e.g. for bladder and swallow control or for paralysis due to different types of palsy).
Mobility
Mobility is likely to be of inherent interest for a paralysed patient. A straight-forward application of BCI is the control of electrical wheelchairs. If classifiable signals can be reliably translated into control sequences for the wheelchair, the paralysed patient regains a certain degree of mobility.
Treatment for attention deficiency
BCI can quantify one’s attention level as measured by EEG waves,
thereby allowing users to employ their attention to play some neurofeedback games. This new treatment provides a safe and interactive tool to guide and
2.9. Using mind control to empower those with disabilities Video
69
regulate the brain from the disorder. This treatment can be effectively used for Attention-Deficit Hyperactivity Disorder (ADHD) without side effects that medication may cause.
Gaming and virtual reality
In addition to medical and rehabilitation applications, there is an increas­ing number of BCI applications for multimedia, such as for simple 2D video games to more advanced 3D video games. There are BCI systems used for nav­igating virtual worlds, and BCI systems used for selecting and/or manipulating virtual objects.
Discussion
Can you think of any other applications for BCI technology?
2.9. Using mind control to empower those with disabilities Video
In October 2016, Switzerland hosted the world’s first Cybathlon, where pilots with severe physical disabilities complete a series of challenges using advanced assistive technologies.
The Cybathlon provides a platform for accelerating the development of new skills and novel assistive technologies that are useful for people with disa­bilities where they need it most - in daily life. Cybathlon athletes not only tack­le challenges, proving their speed and skill using the most up-to-date prostheses and technologies, they also play a vital part in the development of these assis­tive devices.
Dr Mahnaz Arvaneh along with others at The University of Sheffield formed Team Gray Matter, an independent team, who trained to compete in a Brain Computer Interface (BCI) Race.
In this video, we hear from some of the members of the team, including their pilot, Peter Gray, who took some time out from his intensive training at the University to talk to us about his experience so far and his expectations for the system.
Transcript
0:07 We are Team Gray Matter, and at Cybathlon 2016 we will be com­peting against about 20 different teams from universities around the world. Many teams are quite experienced, so it’s a huge challenge for us, as we are
Part 2. Learning from nature
70
starters here. The practical task is to control a computer game avatar to perform actions, to overcome some obstacles, to speed up, or jump, or slide and so on. And this all has to be performed without any actual physical movement. So Peter has to perform mental tasks to control commands for the BCI. For exam­ple, imagining a movement or doing some arithmetic, subtraction, or imagining some visual things - rotating, for example.
1:01 To do mental tasks, and while the user is performing these BCI commands, this is accompanied by some certain types of brain activity, which we can pick up with particular equipment, with electroencephalography, and then use signal processing and lots of data processing and analysis to interpret this activity into some particular control command. This is actually a racing
game, so it’s similar to a lot of apps out there for iPads and iPhones at the mo­ment, but we’re going to be in a virtual race. There’s going to be four avatars lined up and you’re going to get to see not only how you’re performing, but
how well your opponents are performing at the same time.
1:48 It’s, sort of, four states of mind in which I’ve got to train, one of them being neutral, so not triggering any commands and not triggering what
they call artefacts, which is me maybe moving or spasming. It’s very sensitive.
And then the three action commands, I call them, the three different obstacles
we’re dealing with. So, for a purple obstacle, I would imagine reaching for a
chocolate bar with my left hand. Trying to associate the command to the left side of my body and make it an enjoyable experience, something which might make me smile and therefore be detectable in a different way to the next com­mand, which is for the turquoise obstacle.
2:26 It’s almost like an aggressive tackle or an aggressive punch to
somebody’s face. It’s a stark contrast to the nice, smooth, enjoyable reaching
for a chocolate bar. And then the third and final command being something completely different again, which is mental arithmetic. So, doing maths and subtractions in my mind just to use the part of the brain which deals with
maths, and therefore makes it identifiable as a command. It’s my friend Ivan, who’s also the team leader. He set about looking on the internet what was
available out there since my accident. I was previously quite an active, sort of outgoing guy, and Ivan looked from the point of view of getting me fishing again, believe it or not.
3:06 And that’s how he stumbled across Cybathlon and a way for people
in my position to still compete using assistive technology. It’s enjoyable, in