WEBVTT 00:00:00.017 --> 00:00:06.557 This is the convergent science network podcast, 00:00:08.237 --> 00:00:13.057 leading researchers in the domain of neuroscience brain theory and technology 00:00:13.057 --> 00:00:16.777 are interviewed by paul verscher and tony prescott, 00:00:21.257 --> 00:00:27.377 so this is tony prescott for the convergent science network podcasts from the 00:00:27.377 --> 00:00:31.677 barcelona Summer School on Cognition, Brain and Technology in 2011. 00:00:32.137 --> 00:00:37.357 And I'm here with Housang Hu from the University of Essex Department of Computer 00:00:37.357 --> 00:00:41.817 Science and Electronic Engineering, who is one of the speakers that we have 00:00:41.817 --> 00:00:45.217 this week on the general topic of biomimetic robots. 00:00:45.757 --> 00:00:51.177 So, Housang, can you tell me a little bit about your background and how you became a roboticist? 00:00:53.097 --> 00:01:00.077 I came to this country, the UK, 25 years ago. 00:01:00.417 --> 00:01:03.997 Before that, I was working at a Chinese university. 00:01:05.357 --> 00:01:13.277 My major is automation. When I joined Oxford University, I worked with Professor 00:01:13.277 --> 00:01:15.637 based at McBrady, I started doing robotics. 00:01:15.977 --> 00:01:19.197 So that will be back to 1987. 00:01:21.157 --> 00:01:28.457 So I started working with industry, try to produce the new generation of AGV, 00:01:29.237 --> 00:01:33.157 for manufacturing of the industry. 00:01:34.517 --> 00:01:43.537 So started from there, and then I developed my interest in my research in the 00:01:43.537 --> 00:01:48.537 biomimetics or biologically inspired robotics. 00:01:49.257 --> 00:01:55.277 So you have some background in industrial robotics, but your work in recent 00:01:55.277 --> 00:01:59.817 times has been more directed towards service robotics and field robotics. 00:02:00.197 --> 00:02:05.837 So tell me about your interest in field robotics. How did you get involved in underwater robots? 00:02:06.397 --> 00:02:13.857 Yeah, underwater robots is one kind of field robotics. And I'm also working 00:02:13.857 --> 00:02:19.857 on the wheeled and tracked robots and the flying robots as well. 00:02:20.757 --> 00:02:28.157 The reason I got into the underwater robots is because we have an industry approaching 00:02:28.157 --> 00:02:36.797 to us and trying to create robotic fish for the Linden Aquarium. 00:02:36.797 --> 00:02:45.637 Because by law they have many fish species cannot be displayed in the aquarium environment. 00:02:46.157 --> 00:02:54.517 So they want to create a robotic equivalent species to demonstrate how. 00:02:55.475 --> 00:03:00.955 These fish species can swim in the sea, in the aquarium environment. 00:03:01.595 --> 00:03:05.735 So you started off really building robot fish for exhibits. Yes. 00:03:06.415 --> 00:03:10.015 But you've started now moving them out into the ocean? 00:03:10.675 --> 00:03:18.635 Yeah, just the last two years we have support from the EU and FP7 framework, 00:03:18.975 --> 00:03:24.515 and we try to develop robotic fish to patrol ports. 00:03:25.475 --> 00:03:34.255 And monitoring the ship oil leaking problem and also other kinds of pollution 00:03:34.255 --> 00:03:38.495 may cause damage to the sea life. 00:03:39.475 --> 00:03:45.955 So this is a three-year project. We work with a number of partners in the EU, 00:03:46.475 --> 00:03:55.115 like BMT and Irish partners and also tennis and also Spanish partner as well. 00:03:55.815 --> 00:04:01.835 So what would be the advantage in open sea of a robot fish compared to a conventional 00:04:01.835 --> 00:04:03.635 submarine-style robot? 00:04:04.648 --> 00:04:13.028 The fish, as we know, first thing is swimming peacefully without disturbing the environment, 00:04:13.848 --> 00:04:20.328 because the submarine or ship-like, they use the thrust to generate a lot of 00:04:20.328 --> 00:04:25.248 turbulence in the the water, which is no good for sea life, 00:04:25.548 --> 00:04:34.608 and also it may be disturbing pollution area as well. 00:04:34.968 --> 00:04:47.668 So fish is more peaceful and also more flexible in many of the narrow pathways, for example. 00:04:47.988 --> 00:04:52.688 And what about energy efficiency? Is it similar or better than a submarine? 00:04:53.028 --> 00:05:04.348 Yeah, as we know, the submarine type of the man-made ships' efficiency is up to 60%. 00:05:05.028 --> 00:05:13.768 But the fish, theoretically, is higher than the man-made vehicle because after 00:05:13.768 --> 00:05:16.068 millions of years evolution, 00:05:16.548 --> 00:05:22.268 fish is an excellent swimmer in the water. 00:05:22.768 --> 00:05:31.108 They have, of course, the body shape and make the fish swimming much efficient. 00:05:31.408 --> 00:05:40.988 And also, So the muscle movements also make the fish is very fearful as well. 00:05:41.248 --> 00:05:47.268 So you're copying the streamlined shape of the fish and you're also copying 00:05:47.268 --> 00:05:51.528 the type of movement it's making in order to propel itself. 00:05:52.328 --> 00:05:58.388 Is the robot biomimetic in any other way, for instance, in terms of the components 00:05:58.388 --> 00:05:59.808 that you're using to build it? 00:06:01.687 --> 00:06:04.727 Within the robot, are you using conventional motors? 00:06:05.207 --> 00:06:13.027 Yes. Ideally, we hope many researchers actually start using some kind of artificial 00:06:13.027 --> 00:06:16.887 muscle to try to mimic the real muscle movement of the fish. 00:06:17.247 --> 00:06:22.367 But at the moment, artificial muscle is not mature enough for us to use it. 00:06:22.367 --> 00:06:31.767 So what we did, we used normally traditional conventional motors to drive the fish movements. 00:06:32.307 --> 00:06:37.387 And what about the pattern generation systems that are generating the swimming movement? 00:06:37.687 --> 00:06:40.547 Yes, that's what we call the central pattern generator. 00:06:40.907 --> 00:06:42.927 And that's generated by software. 00:06:43.787 --> 00:06:50.127 And after we actually have used the camera to capture the real fish movements, 00:06:50.127 --> 00:07:00.207 We analyze the movements and then to decide the phase difference because the real fish use muscles. 00:07:01.207 --> 00:07:05.667 So muscles can be considered as continuous actuators. 00:07:06.467 --> 00:07:10.827 But we use discrete motors. Motors have different sizes. 00:07:11.347 --> 00:07:17.067 Even if we choose a small size, we only can accommodate four or five or six 00:07:17.067 --> 00:07:24.207 motors in one fish. The fish we built is about 60 cm in total. 00:07:24.587 --> 00:07:30.587 So half the body with the four or five motors, that's what we call discrete 00:07:30.587 --> 00:07:36.347 joints, to actually generate the S-movement of the real fish. 00:07:36.707 --> 00:07:40.507 Okay. So is it a segmented body? Yeah, segmented body. 00:07:40.827 --> 00:07:44.047 And each segment is actually one motor. Right. 00:07:44.307 --> 00:07:49.487 And what about the fins? How many of the fins are you replicating, 00:07:49.587 --> 00:07:50.747 and how important are they? 00:07:50.927 --> 00:07:56.807 Yeah, the fins actually have the main dorsal fin, for example, 00:07:56.807 --> 00:08:01.727 which generates the proponent force for our fish. 00:08:02.027 --> 00:08:06.987 And of course, fish also have anal fins, have the bacterial fins, 00:08:07.287 --> 00:08:16.327 and try to make sure the balance of the fish. So you'd be using those for steering and balance? 00:08:16.407 --> 00:08:20.447 Yeah, that's only for the stationary, static balance. 00:08:21.007 --> 00:08:27.147 But once the fish is in motion, we really use the motor to drive it. 00:08:27.427 --> 00:08:28.787 What about sensors? What kind 00:08:28.787 --> 00:08:34.107 of sensors do you use? We have a number of sensors inside our fish body. 00:08:34.327 --> 00:08:41.667 One is a gyroscope to actually measure the a posture of the fish in the water 00:08:41.667 --> 00:08:43.047 to maintain the balance. 00:08:44.187 --> 00:08:50.807 We also had a force sensor to measure the depth in the water to decide. 00:08:52.838 --> 00:09:00.238 3D swimming, and also we had an accelerometer to measure the speed of the fish 00:09:00.238 --> 00:09:02.978 as a kind of odometer, odometry. 00:09:03.318 --> 00:09:10.878 We also have obstacle detection sensor infrared to detect anything in front 00:09:10.878 --> 00:09:13.458 of the fish in order to avoid it. 00:09:13.998 --> 00:09:22.678 Of course, we had a voltage monitoring sensor to see whether the energy is enough or not. 00:09:22.898 --> 00:09:27.038 If the energy is not enough, we're going to float on the surface. 00:09:27.338 --> 00:09:29.318 Right. Just in case. 00:09:30.338 --> 00:09:34.698 Okay. So how do you control your height within the water? 00:09:35.218 --> 00:09:40.038 To control the height in the water, as we know, real fish have bladder. 00:09:40.398 --> 00:09:43.558 They can be stationed in one level without motion. 00:09:44.298 --> 00:09:47.738 So we mimic the bladder function, 00:09:47.738 --> 00:09:53.578 use the water tank inside of the fish by pumping into the water or pumping out 00:09:53.578 --> 00:10:00.258 of the water to actually change the weight of the fish, change the buoyancy. 00:10:01.238 --> 00:10:07.858 So this is only statically, but dynamically, you want to change the level of 00:10:07.858 --> 00:10:09.998 the fish, go up or go down. 00:10:10.138 --> 00:10:16.058 We actually use one motor to drive the central gravity towards the head or towards 00:10:16.058 --> 00:10:21.198 the tail. in order for the fish to swim up or swim down. 00:10:21.358 --> 00:10:28.658 Very, very agile and very, very speedy operation. 00:10:29.758 --> 00:10:34.118 So I'm thinking about pollution monitoring. So I can imagine that you might 00:10:34.118 --> 00:10:38.558 just have a boat and then you could trail something out of the side of the boat on the end of a line. 00:10:38.678 --> 00:10:44.418 How would having a robot that's submerged be better than other ways of doing that? 00:10:46.598 --> 00:10:54.098 Yeah, because it depends on the depth of the water you are going to monitor. 00:10:54.118 --> 00:10:58.338 For example, we have the port in Spain. 00:10:59.298 --> 00:11:06.478 It's called Gijón, which has a depth of 30 meters. 00:11:07.638 --> 00:11:13.658 So the pollution, if we use the ships on the surface, surface may not be able 00:11:13.658 --> 00:11:20.058 to detect anything beyond 10 meters or seabed, for example. 00:11:20.318 --> 00:11:27.238 So to use the fish, we can automatically change the depth going to the. 00:11:28.253 --> 00:11:33.293 Seabed to actually detect anything, pollution is there or not, 00:11:33.473 --> 00:11:37.473 especially we want to search the source of the pollution. 00:11:38.053 --> 00:11:43.473 So once you put your fish in the ocean, how does it know where to go? 00:11:43.893 --> 00:11:50.413 We actually use the structured environment. We put a buoy surrounding the port 00:11:50.413 --> 00:11:55.293 to send out the ultrasonic signal, 00:11:55.533 --> 00:12:02.693 sonar signal, and the fish has the receiver on board to receive the signal. 00:12:02.993 --> 00:12:04.413 This is kind of like underwater GPS? 00:12:04.953 --> 00:12:09.853 Yes, exactly. Right, okay. So it knows where it is, but you're never worried 00:12:09.853 --> 00:12:10.953 about that it won't come back? 00:12:11.593 --> 00:12:22.353 Oh yes, we actually have to prevent in case of underwater GPS malfunction, fish may get lost. 00:12:23.253 --> 00:12:27.573 So in that case, fish going to floating on the surface, we use GPS. 00:12:27.933 --> 00:12:30.833 Right, so you can use the real GPS, satellite. 00:12:30.833 --> 00:12:36.433 Yes, because the antenna will be floating over the surface of the water, 00:12:37.253 --> 00:12:43.793 then GPS signal can be received, and then we can navigate back to the home position. 00:12:44.333 --> 00:12:47.193 So how close do you think we are 00:12:47.193 --> 00:12:52.113 to having a technology that might be commercialized for this sort of use. 00:12:52.333 --> 00:12:57.713 Yeah, this project we are running now is third year. 00:12:58.753 --> 00:13:04.093 So we hope we're going to test this in the port next year, next May. 00:13:05.273 --> 00:13:11.593 And after that, I think we need a couple more years to refine it and improve the performance. 00:13:12.593 --> 00:13:15.373 I imagine maybe three or four years 00:13:15.373 --> 00:13:21.733 time From now on, we could have some commercial products on the market. 00:13:22.053 --> 00:13:25.933 And what other application areas might there be besides pollution monitoring? 00:13:26.453 --> 00:13:37.333 And also security. For example, some of the coastlines, they want to secure for different reasons. 00:13:37.673 --> 00:13:45.233 So we can use this kind of fish technology actually to detect any illegal ships. 00:13:45.733 --> 00:13:49.873 Or any illegal animation. Right, yeah. 00:13:51.182 --> 00:13:55.882 So it could be part of the Coast Guard, or it could be part of the Customs and 00:13:55.882 --> 00:13:58.002 Excise. Exactly. Those kinds of things. Yes. 00:13:58.262 --> 00:14:03.242 Okay, so changing subjects a little bit, I know that you have an interest in 00:14:03.242 --> 00:14:07.222 building assistive robots for people with disabilities. 00:14:08.082 --> 00:14:11.762 Can you give me an example of the kind of projects that you're doing there? 00:14:12.342 --> 00:14:21.602 Yes. This is to try to help disabled and aging population because when people 00:14:21.602 --> 00:14:23.462 are getting old or people are disabled, 00:14:23.842 --> 00:14:31.782 they have difficulty to move around or independent life has been out of question. 00:14:33.162 --> 00:14:41.222 So what we try to help is mobility, to help these people move around, 00:14:41.482 --> 00:14:43.562 integrate with society, 00:14:44.302 --> 00:14:52.882 and they can see the doctor if they want, they can see their relatives or friends go out if they want. 00:14:53.082 --> 00:14:57.762 So that's the main purpose. As we know currently, 00:14:58.142 --> 00:15:07.662 commercial wheelchairs are only driven by joysticks, and many users have difficulties 00:15:07.662 --> 00:15:12.422 to navigate with joysticks, even go through the doorway. 00:15:13.162 --> 00:15:22.282 For example, users suffer from Parkinson's disease, their hand is shaking many 00:15:22.282 --> 00:15:26.602 times, sometimes they may not be able to use joysticks. 00:15:27.002 --> 00:15:33.922 And some of disabled people, they may be suffering no hand, for example. 00:15:34.142 --> 00:15:35.942 It's also difficult to use. 00:15:36.482 --> 00:15:40.042 So in that case, we think about the other means. 00:15:40.342 --> 00:15:47.502 For example, use the voice, use the gesture, use the muscle signal, even for. 00:15:49.112 --> 00:15:56.732 Disabled users. We use like Hawking, Professor Hawking, and cannot move any limbs. 00:15:57.032 --> 00:16:07.472 And then, if their brain functions well, we use EEG signal to control wheelchair movements. 00:16:08.012 --> 00:16:12.152 So you want to record EEG, sort of an electroencephalogram. 00:16:12.732 --> 00:16:20.092 So this is electrical activity from nerves, which you record by placing sensors on the skin. Yes. 00:16:20.432 --> 00:16:23.892 And whereabouts would you put the sensors? 00:16:24.352 --> 00:16:30.152 We actually put the sensors in the skull, on the motor cortex area. 00:16:30.692 --> 00:16:34.712 And the users actually can just imagine. 00:16:36.052 --> 00:16:44.292 His hand movements, his or her neck movements, then generated control commands 00:16:44.292 --> 00:16:48.012 for the wheelchair to be controlled. 00:16:48.532 --> 00:16:52.872 So this is the part of the brain, the motor cortex, which is anyway involved in motor control. 00:16:53.112 --> 00:16:58.132 Yes. And so you're hoping that by recording through the skull some of the electrical 00:16:58.132 --> 00:17:02.792 activity, you may be able to distinguish when the user wants to, 00:17:02.812 --> 00:17:05.312 say, turn left or wants to turn right. 00:17:05.432 --> 00:17:08.992 That's right. So, I mean, does this work? 00:17:09.132 --> 00:17:12.632 I mean, is it possible really to read these things from that kind of brain activity? 00:17:13.032 --> 00:17:16.552 Right now, we actually have some results. 00:17:17.052 --> 00:17:24.312 Although it's not real-time, and sometimes the success rate is maybe up to 70%. 00:17:24.312 --> 00:17:31.092 So, we think the key problem is the sensor. Yeah. 00:17:31.764 --> 00:17:41.044 Right now, we use non-invasive electrodes on the skull, and the signal is very weak. 00:17:41.424 --> 00:17:45.824 So we have to improve the sensor technology. 00:17:46.224 --> 00:17:52.304 We hope by next few years, if the EEG sensor can be further improved, 00:17:52.624 --> 00:17:55.484 then we can use such technology. 00:17:56.044 --> 00:18:04.584 This is one thing. And also, another thing is we think maybe multiple modality, 00:18:04.804 --> 00:18:07.904 not just EEG, and also facial, 00:18:08.264 --> 00:18:13.344 also other biosignal can be held. 00:18:14.784 --> 00:18:17.564 Combined with EEG if possible. 00:18:18.084 --> 00:18:23.164 What other kind of signal are you thinking of? Like facial, emotion, 00:18:23.684 --> 00:18:32.584 facial, eye movements, and mouth movements, simply that clue also can reflect people's needs. 00:18:32.944 --> 00:18:38.844 Okay, so if I'm a wheelchair user and I'm wired up with these electrodes on 00:18:38.844 --> 00:18:44.404 my head, what do I have to do to make the wheelchair move? 00:18:44.524 --> 00:18:47.824 Is it enough that I think I want to go left and then it will turn? 00:18:47.924 --> 00:18:52.884 Or do I have to learn the relationship between my thought patterns and what 00:18:52.884 --> 00:18:56.164 the wheelchair does? Yes, we do need training. 00:18:56.624 --> 00:19:05.384 We need every user to train to generate imaginary signals. 00:19:05.824 --> 00:19:13.664 For example, a user can think about using the right hand, or using the left hand, or using the legs. 00:19:14.104 --> 00:19:20.184 So we we can generate the electric pattern on the transducers. 00:19:21.144 --> 00:19:28.344 Then we need to train our, we use the neural network and we need to train all the, 00:19:29.262 --> 00:19:32.342 all the parameters so you train so 00:19:32.342 --> 00:19:35.662 you record uh these eeg patterns yeah 00:19:35.662 --> 00:19:40.162 for people thinking about a particular kind of movement yeah and you train a 00:19:40.162 --> 00:19:45.342 neural network with those patterns to distinguish left thoughts about moving 00:19:45.342 --> 00:19:51.082 left from thoughts about moving left right yeah and um so roughly speaking um 00:19:51.082 --> 00:19:55.522 how much training data would you need i mean from a person would you need to several 00:19:55.602 --> 00:19:57.962 hours of EEG or could you do it quite quickly? 00:19:58.662 --> 00:20:04.602 We normally doing several hours. Right. We normally doing several hours. 00:20:04.942 --> 00:20:09.962 And with that person having trained on their data, would it be stationary or 00:20:09.962 --> 00:20:13.902 would you need to train again if you came back two weeks later, 00:20:14.062 --> 00:20:16.402 would it be different or would it still work? 00:20:16.682 --> 00:20:24.042 Yeah, that's a good question. Actually, people, they have diverges because they are mental. 00:20:25.582 --> 00:20:30.322 Condition. For example, if a user sleeps very well last night, 00:20:30.522 --> 00:20:33.862 so today they may function very well. 00:20:34.082 --> 00:20:41.362 If they don't sleep very well last night, they may not get a useful signal today. 00:20:41.942 --> 00:20:44.042 So in that sense, our, 00:20:46.213 --> 00:20:50.653 algorithm have to be more adaptive. So we call it online learning. 00:20:50.973 --> 00:20:56.333 So what we want to do, not just offline, and also we want to online continuously 00:20:56.333 --> 00:21:00.753 change the parameters of our neural network, 00:21:00.893 --> 00:21:05.993 for example, to adapt to new stages of the mental stage of the users. 00:21:06.573 --> 00:21:11.333 Okay, so where have you got to in terms of testing these? 00:21:11.373 --> 00:21:15.693 Have you got people driving around with with wheelchairs and controlling them 00:21:15.693 --> 00:21:19.193 just with their thoughts? Have you got some prototype results now? 00:21:19.493 --> 00:21:24.353 Yes, we do, we did. And we have, right now it's health subject. 00:21:24.653 --> 00:21:28.313 It's all, in fact, our PhD students. 00:21:29.613 --> 00:21:37.913 So next stage we hope to get some disabled people or trial in the real users. 00:21:38.333 --> 00:21:41.713 So how good are the students then? And can they, for instance, 00:21:41.853 --> 00:21:45.733 drive the wheelchair through a narrow doorway? 00:21:46.433 --> 00:21:55.253 Yeah, and they can navigate in the indoor environment in general, no problem. 00:21:55.533 --> 00:22:03.573 Without bumping into things? Yes, because the control system has prevented bumping into anything. 00:22:04.393 --> 00:22:10.153 Use the sensors. Oh, so there are sensors on the wheelchair as well to prevent collisions. Yes. 00:22:10.653 --> 00:22:15.913 So that you have a backup in case the brain reading system is not working very well. Exactly. 00:22:16.313 --> 00:22:22.033 So the wheelchair is independently sensing the environment and making some decisions 00:22:22.033 --> 00:22:24.293 about when to stop or when to go. 00:22:24.473 --> 00:22:28.873 And then really what you're doing is reading the person's brain state in order 00:22:28.873 --> 00:22:32.593 to bias what the wheelchair is going to do. 00:22:32.673 --> 00:22:39.053 Yeah, either EEG signal or muscle EMG signal. and also maybe other signals like 00:22:39.053 --> 00:22:45.853 voice, like hand gesture or head gesture, use the vision-based. This is all... 00:22:46.772 --> 00:22:50.852 We call the hands-free control. It's on top of the navigation system. 00:22:51.272 --> 00:22:57.152 So all this human intention, biosignal I mean, to detect by the sensors, 00:22:57.432 --> 00:22:59.232 use to control the wheelchair motion. 00:22:59.592 --> 00:23:04.752 But at the bottom of the wheelchair control system, we have safety control system 00:23:04.752 --> 00:23:10.332 there with the laser scanner, with the ultrasound sensors, 00:23:10.812 --> 00:23:15.452 they can monitor surrounding objects of the wheelchair. 00:23:15.452 --> 00:23:21.352 If any object is close to the moving direction of the wheelchair, 00:23:21.592 --> 00:23:28.632 the human intention commands will be not executed because of a safety issue. 00:23:29.052 --> 00:23:33.452 Okay, so you're fairly confident that the wheelchair can move around safely 00:23:33.452 --> 00:23:35.712 even if the person falls asleep. 00:23:36.112 --> 00:23:41.992 Sure. Yeah, exactly. No, I'll give the wrong commands because sometimes that happens. 00:23:42.612 --> 00:23:50.452 So I guess when you want to use this in earnest, gluing electrodes to people's 00:23:50.452 --> 00:23:54.032 skulls isn't going to be very popular with wheelchair users. 00:23:54.252 --> 00:23:58.332 Have you got a plan for how to make this interface more easy to use? 00:23:58.652 --> 00:24:02.192 Yes, what we're trying to do is we try to use wearable sensors. 00:24:03.212 --> 00:24:11.972 To put the sensor on the head, put the sensor on maybe a glass, 00:24:12.252 --> 00:24:18.592 some sensors, and also put the sensors on the body or in the clothes. 00:24:18.872 --> 00:24:30.152 Make the sensors wearable, portable, and compact, so people wouldn't... 00:24:35.429 --> 00:24:44.809 Wouldn't, how to say, people wouldn't bother by this kind of technology. 00:24:45.389 --> 00:24:52.449 Right. Does the technology already exist for, for instance, voice control of wheelchairs? 00:24:53.729 --> 00:25:00.009 The voice control wheelchair has been developed for several years now, 00:25:00.169 --> 00:25:11.069 but not much use in the real situation because the background noise may cause problems, 00:25:11.269 --> 00:25:17.029 and also people with different accents also need training as well. 00:25:17.409 --> 00:25:24.569 So I can imagine in the future, we have to combine different modalities with 00:25:24.569 --> 00:25:28.969 the voice, for example, voice with visual signal. 00:25:28.969 --> 00:25:35.129 For example, if we use a microphone to catch up the human voice, 00:25:35.509 --> 00:25:40.469 at the same time we use the camera to catch up the lip motion, 00:25:41.589 --> 00:25:49.409 then we can confirm the user's comments by tracking the lip motion and also 00:25:49.409 --> 00:25:53.249 the voice to actually make it more robust. 00:25:54.589 --> 00:26:02.409 Even background noise is very loud. So you're confident that these kind of controlled 00:26:02.409 --> 00:26:07.989 wheelchairs are going to be something that people will be using in maybe a few years' time? 00:26:08.709 --> 00:26:17.329 That's exactly it. Recently, we have a charity that would like to invest in this area. 00:26:17.569 --> 00:26:23.509 So we have a plan to commercialize in three or four years' time. 00:26:23.509 --> 00:26:30.489 So moving to broader questions about robotics, what do you see as the really 00:26:30.489 --> 00:26:34.229 big challenges in the field in the next few years? 00:26:34.269 --> 00:26:38.049 The things that are maybe the bottlenecks that are stopping us from building 00:26:38.049 --> 00:26:40.369 more intelligent robots, more useful robots? 00:26:40.869 --> 00:26:47.569 Right. I think the key challenge is the sensing of the environment, 00:26:47.989 --> 00:26:50.549 sensing of the human intention. 00:26:51.649 --> 00:27:00.089 As we know, the robots is the integration of the sensor technology and also 00:27:00.089 --> 00:27:05.409 AI and the learning ability, of course, actuation side as well. 00:27:05.849 --> 00:27:12.949 I think the 21st century is a robotic century, means the robots can bring the 00:27:12.949 --> 00:27:16.769 benefit to our society to improve our quality of life, 00:27:17.749 --> 00:27:22.169 exactly like a computer technology and change our lifestyle, 00:27:22.409 --> 00:27:25.069 change our working style in last century. 00:27:25.389 --> 00:27:34.629 So, but the key problem here now is how we created an interface between technology. 00:27:36.006 --> 00:27:41.126 Human and robots. And right now, there's still a bigger gap. 00:27:41.686 --> 00:27:46.006 All the robots have to be used by the people with training. 00:27:46.666 --> 00:27:52.966 And so if we want the robots going to individual home, 00:27:53.186 --> 00:28:01.706 the robots have to be easy to use without training on the programming side. 00:28:01.706 --> 00:28:08.546 Otherwise, still 80% of users wouldn't be able to use robots. 00:28:10.246 --> 00:28:16.006 To finish off then, what would be a prediction you might make about the future 00:28:16.006 --> 00:28:21.146 of our society and how we might be using robots in say 10 or 20 years from now? 00:28:21.386 --> 00:28:25.846 What can you think of an example of one big change that robotics is going to 00:28:25.846 --> 00:28:27.906 bring about in our society? 00:28:28.786 --> 00:28:32.146 I think, as I said earlier, 00:28:32.426 --> 00:28:38.766 the 21st century is a robotic century, and I can see a lot of different forms 00:28:38.766 --> 00:28:46.726 of robots, of different kind of robots going into our homes, 00:28:47.646 --> 00:28:50.466 and hospitals and everywhere, 00:28:51.366 --> 00:28:58.886 especially for the dangerous situation like such a rescue work or like a nuclear 00:28:58.886 --> 00:29:00.466 disaster, this kind of things. 00:29:01.686 --> 00:29:05.286 And also many of the housework as well. 00:29:06.172 --> 00:29:15.012 I can imagine in 20 years' time, many of the robots are going to be portable 00:29:15.012 --> 00:29:19.772 and also going to be wearable. 00:29:19.772 --> 00:29:25.212 For example, in the future, our soldiers in the battlefield may have some kind 00:29:25.212 --> 00:29:34.632 of GPS system located themselves, and have been tracked by the army for safety purposes. 00:29:34.632 --> 00:29:42.792 Purpose, and also many of the facilities at home with the robotic technology. 00:29:43.312 --> 00:29:50.632 For example, even like a fridge, they can tell users, the milk is going to be 00:29:50.632 --> 00:29:55.772 finished, you have to bring milk from the supermarket. 00:29:56.352 --> 00:29:59.292 Or maybe get the robot to bring milk from the supermarket. Yeah, 00:29:59.332 --> 00:30:08.092 and also, remember, Remember, we can see some kind of iPhone is kind of robots 00:30:08.092 --> 00:30:10.512 as well, if that's the meaning. 00:30:10.752 --> 00:30:14.932 Right. Well, thank you very much for talking to us. So this is Tony Prescott. 00:30:14.992 --> 00:30:16.772 I've been talking to Hu-Seng Hu. 00:30:16.932 --> 00:30:19.452 Thank you. Thank you. Thank you very much. 00:30:23.172 --> 00:30:28.712 The CSN podcast was produced by the Convergent Science Network of Biometrics 00:30:28.712 --> 00:30:35.532 and Biohybrid Systems, a project funded by the European Sevens Research Framework Programme. 00:30:36.652 --> 00:30:41.972 For more interviews, recorded lectures or upcoming conferences in the field 00:30:41.972 --> 00:30:48.212 of biometrics and biohybrid systems, go to csnnetwork.eu. 00:30:48.240 --> 00:30:56.240 Music. 00:30:48.532 --> 00:30:50.392 And thank you for listening.