 Presenting on a Continuum Manipulator Refinement and Control, my name's Sam. My name's Kayla and our third partner is currently in Europe right now. So she can't make it. So what is assistive technology? Assistive technology is an item, piece of equipment, software program, or productive system that increases, maintains, or improves the functional capabilities of people with disabilities. There's a wide range of assistive technologies that go from learning systems mobility, vocational aids, computer access, and so forth. So assistive tech can be geared towards those who struggle to perform daily activities independently and without assistance. So some activities that they might need help with would include bathing, brushing your teeth, changing your clothes, eating or drinking, or enabling mobility. So with riders in use of reliance on health care, there has been a push to have robotics in assistive technology. And currently there are three different types that are used. There's the social assistive robot that's used with elderly patients to help improve their physical and mental well-being. So they have responsibilities and it helps. So this is PARO, which is one form of social assistive robot. There's also therapy robots that are used to assist with physical therapy. Another example is the MIT MANIS, which the user will guide it in most cases, but if their arm isn't moving, the robot will not jump to create the movements to help with physical therapy. And then there are also robots that replace limited limb movement, so if your arms don't work as well, and things like that. So rigid-linked robots are examples of those that can help with limited limb movement. So these two, the i-arm and the jaco arm, are the two most popular on the market today. They're both rigid-linked robots that are seven degrees of freedom. They're both wheelchair-mountable, and they use encoders and or torque sensors to help guide precise movement. And they're a bit expensive. The jaco arm can go for about $35,000, and insurance does not always cover this. So oftentimes the load will be put onto the consumer. So what is a rigid-linked robot? A rigid-linked robot has rigid and flexible segments, which are connected by joints, which is where your actuation comes from, which the number of joints is your number of degrees of freedom. They have precise control. Both sensors are limited based on their movement on joints. They have limited degrees of freedom. They're rigid and hard, so if they're running to someone, they can cause injury. And then they can be very costly with all the sensors that are needed and controls to help them be safe around humans. So there are definitely some pros and cons to using rigid-linked robots. There is another type of robot that's been gaining popularity in many fields of research and application, and that is the continuum robot, or the continuum manipulator, which is popular in research like medical fields and outer space assistance. So continuum manipulators are robots that do not have the same rigid structure and discrete joints that rigid robots have. So if you think of the movements of elephant trunks, snakes, or octopus tentacles, then that's kind of the movement that they simulate. So a benefit of this type of manipulator is their extended capabilities of connecting with their surroundings. So for example, the rigid-linked robot can only use their end-effector or the example of the jaco arm, their hand, to interact with their surroundings. A continuum robot may be able to use not only their end-effector, but also their entire backbone to interact with the environment. One of the pros of this type of robot is that they are naturally highly compliant and elastic. So this increases flexibility, possibly range of motion. Also, it makes them more suitable for dynamic and challenging environments. So this does come with a bit of a cost. These robots have been difficult to model and control. And additionally, it may be difficult to find the perfect material that's flexible yet strong enough to do the ideal activities. And so we were thinking that continuum robots might have a place in assistive tech. So when you think of assistive tech, some of the priorities is that the human is safe and comfortable and able to use the technology. So we were thinking that assistive robots that are continuum manipulators might benefit because they're inherently safer. When you interact with one, it won't necessarily harm you. It may move away with the force or wrap around you. Additionally, like I said before, it is flexible and compliant, and so it may be able to navigate challenging environments better. So things in awkward angles, whether it's people around you, it may be better suited for those types of situations. And additionally, it is potentially more economically efficient. You don't need to spend so much money on the hardware aspect of it. So one research group looked at using a soft robot in a three-hour setting. And you'll notice that the shower robots that are currently on the market don't inherently interact with their user. One example is there's a chair that's a shower robot and parts of it will move away. I'll just spray water on you. So they wanted to design something that could interact more with their user. So their goal was to make a robot that could reach the back area and could apply a soap and potentially scrub that area for their user. This is their setup of their experiment to see how well their design worked and moving because they have tendons and pneumatics located in their robots. They wanted to see how well it got to the different curves and areas. For their control scheme, they used a reinforcement learning. So their agent interacted with the Markov decision process. The robot got awards. The closer it got to different tile lengths. So they had tiles set up through their area. So as there's a sphere around each tile, and as it got closer, the amount of reward for a computer got higher. It was reliant though on an off a vision track system so that way they could tell where they were at in their system. And once after their testing, they noticed that the position of accuracy for about 12 points of motion was 0.79 centimeters. Give or take about 0.18 centimeters. So we worked off of a previous project of the continuum robot. That robot was double-submitted. It had the backbone was created, was made of a plastic low-density polyethylene backbone. It is tendon-driven. So they used a braided fishing line, which routed through 3D disks that were evenly spaced along the backbone, and actuated by eight stepper motors. So it was designed with eight degrees of freedom, but in theory does have infinite degrees of freedom, and the end effector is an electromagnet. And so the user was able to control this robot with three different types of control schemes. So you could either use two joysticks to control different segments of the robot. You could use a single joystick, which you could change the mode to control different parts of the robot or the robot as a whole. Additionally, you could use a compensative mode, which would have the robot attempting to decouple the top and bottom segments as you drove it. And then there was also a homing feature, where the motor rotations were used to bring the robot to an upright position. So the goal of our project is to design a robot that can be automated. So we didn't want to use vision feedback because it would increase the cost a lot. Additionally, it would hinder the mobility of the robot and add additional material for the robot. We were thinking that sensors are more affordable and can also provide accurate information. And we consider these factors because our end goal is to be able to open source the code and the materials list so that people from home can build the robot and then use it for themselves without the high costs that come with everything robot. With last year's project, they had some limitations based on the design that we focused on the ones that affected our goal of automation. A big one was their backbone material. There was deformation in this section down here that prevented it. So instead of being vertical, it had a slight curve to it. So some of the driving operations and homing feature didn't work as well as they were hoping to work. There's also mechanical coupling. So with continuum robots, you should be able to operate the top and bottom sections as separate sections. So you can move the top without moving the bottom. But they noticed they had two sections on their backbone. So with the bolt in the middle, some of the fourth moment was transferred down, which prevented the coupling. But also with having the tendons running through the bottom section that controlled the top section, some of the force was placed into the bottom section so the robot would move it in a single piece instead of allowing for the two disjoint sections. Our biggest focus for the summer to get to automation was replacing this backbone material. Our first election we looked at using was a nitinol, which is a nicotitinium alloy. This is due to its property of super elasticity, which comes from when at room temperature. It's a nearest transition temperature. So when strain is added, it will actually go from an austenite material structure into a martensite. And then once the strain is removed, it will go returned back to the austenite deformed material structure so it will be able to return to its undeformed position. However, we noticed that during operation, our three millimeter rod, this bend was created during an extreme bend that occurred in the motion and we were unable to even, after the robot was disassembled, just pull it out. And we decided that having that, even if it was an extreme bend, if our users were able to get it to those bends, then the material would be deformed so it was not valid for long-term usage. We thought maybe a smaller version of the rod would help with that. However, we just noticed it had a lot more deformation along the backbone. So the first original backbone was a low-density polyethylene material. And we decided to go back to that material, but as a single piece, as seen here, to check to see if maybe it's a material selection. So currently, we haven't noticed a lot of deformation down in this section. So we're trying to decide if it's a fatigue issue in the material or if it's a physical design aspect and how just the rod's created and how operates that section will have higher strains that will need to be focused on larger. So with the mechanical coupling, we did notice that when we were able to move the tap a little less slightly when the bolt was removed, but not as much as was expected. So we looked into using cable conduits. Cable conduits allow us to run our tendons through the bottom section, here, up into the top section without the force being applied into the bottom section because they are flexible to allow the curvature that is needed, but they don't compress the tendons like they don't change in that bottom section but only in the top section. And we noticed, as with driving, that there was actually, there's still some coupling since the material selected isn't the greatest and they don't run the whole length, but there's definitely some slight movement that we can occur at the top without the bottom moving. The issues with our material selection is currently they can't elongate, so we weren't able to attach them to the bottom segment. So when we are in rotate, when we are in a bend on the inside of the curve, we notice that the cable conduit will go beneath the desk pulling the tendon down in that area. But like I said, with our elongation on the outer curve, we needed to maintain that to increase that strength. So one way to fix this is there's an accentual spring, so they come in a compressed state when they're naturally relaxed. They can elongate to the size that we need, but they won't be able to go back past the compressed state that they come in. So regarding automation, other consumer robots may be designed to use encoders, optic tracking, or other sensors to gather positional data. We decided to go with motion processing units, otherwise known as MPUs, in order to get data regarding orientation. MPUs have both a gyroscope and an accelerometer in them, so we can get rotational velocity from the gyroscopes and then acceleration raw data from the accelerometer. We went with these sensors for several reasons, like I mentioned before, one that it's cost effective. It's simple to use without requiring a lot of additional equipment, and the acceleration and rotational data seemed feasible to get us what we needed. So we ended up using two sensors in order to keep track of the robot, one at the top of the top section, and then one at the top of the bottom section of the robot. Using the acceleration data, we were able to get tilt angles of the disk that they're situated on, so essentially calculating the angle between the line and the axis. So using x, y, and z acceleration data, we were able to utilize x and y inclination angles. Initially, one of our concerns was drift air, so technically it was possible to get Cartesian coordinate points from the acceleration. However, over time, the noise would accumulate and cause too much air to be useful, and so it would be unstable, unstable to use. But we were able to average out the data coming in and then get reliable x and y angular data, which is why we also went this route. So, you know, we were hoping to use this information to help us automate and drive the robot to the correct, specified place. So additionally though, as a second measure, we wanted to use tendon links to help bring the robot to the correct position. So, you know, given our tilt strategy, there is still room for air, so this was going to be a supplementary measure. The previous project had used the motor rotations in order to automate the robot home to an upright position. However, because the tendons tended to get loose, that wasn't the most accurate method of going about it. So we have tried to solve that by making sure the tendons are taught during movement and during automation, and additionally by utilizing arc length in order to find the tendon links of the desired position and of their current positions. We recently tested this and then found that it didn't exactly go as expected. So the tilt angles, we will yield expected tilt angles upon movement, however, sometimes when the robot rotates, it'll throw those off. So we decided that we did need to include something like rotation angles in order to better understand the orientation. So tendon links is still possible, but with the current use of the inclination angles, it might just throw the robot arm into a different direction than expected. So we're thinking that for the next iteration, if you utilize inclination angles and rotation angles, then you would definitely get a better idea of the orientation, and then it still could be considered to use the tendon links alone to help decide orientation. A whole picture of a robot, which is much easier just to see. So the cable conduits are running. They originally ran the whole length of the bottom section, but with these limit switches here, we noticed they would actually break the limit switches sometimes. So we ran them only this far, which we did notice during driving that they are as efficient in controlling just the top section. So definitely finding a new home for the limit switches so they can run the whole length would be improved. And then our gyros are these right here on the top and middle section of the robot. And then Sam can drive it. Well, is there any questions while we set it up? No, we're ready for questions.