To enable a robot to learn to play ping-pong, the robot must have a perception system that can track the ball to choose its actions (e.g., how to adjust its swing) and evaluate the effectiveness of its actions (e.g., how good was the shot). To accurately track the ball’s trajectory, the system must infer the ball’s position and velocity, which are subject to gravity, drag, and the Magnus effect. To account for this latter effect, prior work has assumed explicit markings on the ball to obtain a point estimate of the ball’s spin; however, doing so renders the balls illegal for use in professional ping pong. In our research, we seek to overcome this limitation in prior work by proposing a model that can be used foe accurately tracking regulation of ping pong balls. Our system consists of Canny edge detection and Hough line transform to detect the ping pong table as well as a Hough circle transform to detect the ball. We then derive a state-transition model within an Extended Kalman Filter (EKF) framework in which the spin is a latent parameter to be inferred by the EKF. By doing so, we can explicitly account for the Magnus effect with regulation ping pong balls. In ongoing work, we are integrating our system with a robot ping-pong player to provide the state and reward signal for training via reinforcement learning.

Feasibility for Proportional Electromyographic Control of a Knee and Ankle Prosthetic
Alex Eichinger-Wiese, University of North Carolina Asheville
Faculty Advisor: Aaron Young, School of Mechanical Engineering

Control of an ankle and knee powered prosthesis has proven to be a difficult task for both the users and researchers of these devices, thus the addition of a direct link between the user’s muscle ac- tivation and the device should improve the response of the device. This pilot study was designed to deter- mine the feasibility of using signals from specific mus- cles for modulating the push off power of the robotic ankle joint. Motion capture and force plate data were collected from N=2 subjects using a Vicon motion capture system, Bertec force-instrumented treadmill, and a Biometrics EMG (Electromyographic) system. Users were asked to ambulate at different speeds and accelerations and to climb various stair heights. A combination of force plate data from the force plates and the Biometric EMG signals were analyzed using signal processing and gait cycle segmentation to create an average EMG signal for every mode of transport. The relationship between the amount of ankle of push-off needed by the user and the EMG signal from one muscle should allow for the creation of a control protocol for the prosthetic device that is more responsive to the user’s needs. The results indicated that there is a possible correlation between EMG signal and ankle push off, but more analysis is needed to show a definite trend. This pilot test will set the ground foundation for developing a more natural and intuitive control, specifically EMG proportional control, on the experimental prosthetic device.

Patient Information Visualization Using Augmented Reality
Brenna Fankell, University of Wyoming
Faculty Advisor: Charlie Kemp, School of Biomedical Engineering

Enabling clinicians to rapidly understand patient information could improve the efficiency and accuracy of patient care. Clinicians might be able to save time on the floor by visualizing patient information through intuitive graphics on an augmented reality display. We present Patient Information Visualization using Augmented Reality (PIVAR), an augmented reality scene understanding method that superimposes a volumetric human model on a patient in bed as a substrate for presenting relevant information. Key areas of interest, such as surgical incision sites, could be displayed as holographic features on the virtual body with pointers to condition descriptors and key metrics. To communicate up-to-date information, our system sends real-time information from information sources to an augmented reality head-mounted display using the robotics middleware Robot Operating System (ROS). Our system uses sensors on the bed to estimate the human pose in real time, which may be useful for presenting the location of a covered wound or incision site. We demonstrate the accuracy of the HoloLens to project a 3D virtual body onto a person lying in a bed by measuring the Euclidean distance between joints of participants with corresponding joints of the mesh. The overall joint error distance was 72 mm. This shows the feasibility of the HoloLens to accurately project a 3D mesh and spatially co-registered patient information over a person lying in bed.

Optimal Human Attention Scheduling Based on Human to Agent Trust Model
Yuki Gao, New York Institute of Technology
Faculty Advisor: Fumin Zhang, School of Electrical and Computer Engineering

In the past, robots were isolated in factories, but as technology advances, robots are stepping out of factories and began to interact with humans. Although technology advanced, robots still cannot take over human labor because robots have not reached human expert level yet. As said, human operators are still important and considered more expert than robots. Human can guide robots and perform Human Robot Interaction (HRI), but there are problems presented. In order to resolve the problems presented, our research objective is to assign priority to robots to resolve conflicts among human attention, to implement commonly used scheduling algorithms in real robot platform, and lastly, compare their performance and determine the best method used. Furthermore, for this research, we implemented different types of scheduling methods like Static Scheduling, Earliest Deadline First (EDF), and Lowest Trust First (LTF) into the Georgia Tech Miniature Autonomous Blimp (GT-MAB). We also ran simulation results for each of the scheduling algorithms to determine which scheduling method is the most optimal out of the three experimented. After comparing all the simulation results, we have concluded that Lowest Trust First is the most optimal out of the three scheduling methods experimented. This result is significant because from the beginning we predicted that LTF will be the most optimal out of the three scheduling methods used for HRI experiments.

Haptic Feedback-Enabled Proprioception of a Soft Robotic Tail
Melanie Hook, Slippery Rock University
Faculty Advisor: Frank Hammond, School of Mechanical Engineering

The objective of this research is to examine the use of haptic feedback, specifically vibrotactile stimuli to the torso, to facilitate the proprioception and human control of a wearable, non-anthropomorphic robotic tail. We aim to determine if utilizing sense of touch via vibrations will eliminate the need for visual data to estimate the tail’s position. A robotic tail was built using a soft continuum design, chosen for its near-constant curvature and under-actuation, resulting in infinite degrees of freedom. The robot was made of a low-density foam and actuated by three cables running the length of the tail. Tension sensors attached to the cables send data estimating the tail’s position to a wearable haptic display consisting of three vibrating motors embedded into a belt. This project has demonstrated that haptic feedback is a promising method for helping the user to sense the general position of a robotic tail. Further work is necessary to determine the threshold for detection of vibrotactile pulses, in addition to optimizing the pattern and duration of stimuli to present the information in the most intuitive way. Overall, this research is an important step to expanding the range of wearable robotic devices to include non-anthropomorphic robots, such as a tail, that have the potential to provide a variety of non-traditional functions.

Novel Ultraviolet-curing Adhesive Gripper Design and Evaluation
Esther Lee, North Carolina State University
Faculty Advisor: Anirban Mazumdar, School of Mechanical Engineering

A gripper that uses an ultraviolet-curing adhesive was designed and prototyped in which the device cures itself to the target object using ultraviolet (UV) emitters, allowing for large amounts of weight to be held by the device. The team’s efforts also revolved around quantifying the load capabilities of the adhesive based on UV emitter cure time. The results of load testing demonstrated that there is a plateau of approximately 30 pounds, beyond which the UV cure time is irrelevant. This load is reached around 75 seconds of cure time. These results indicate that there is a potential to save energy by minimizing cure duration without sacrificing performance. Additionally, the load tests demonstrate the potential for the gripper to hold weights significantly heavier than the prototype itself. Tensile strength across different materials was also tested, and the results showed that load capabilities were largely constrained by the adhesion to the acrylic. Simultaneously, the team built a gripper prototype in assembly with a peristaltic pump for the epoxy, UV emitter arrays, and a DC motor. Testing the prototype on a robotic platform proved that the gripper design is functional and applicable to moving heavy objects.

Hildebrand Analysis for Hexapod Locomotion
Divya McFarland, Penn State
Faculty Advisor: Daniel Goldman, School of Physics

Robots that travel using legs are capable of successfully crossing more complicated terrains than robots that move by other means. However, the flexibility of motion that comes with the presence of legs makes the gait of the robot increasingly complex. Current gait studies are primarily focused on quadruped robots, so there is less information available on robots with additional legs. While many six-legged insects are observed using the alternating tripod gait, it was unknown if this was the most effective way to move. The goal of this study was to evaluate how different gaits affect the forward progress of a hexapod robot. Gaits were defined based on duty factor, leg phase shift, and the presence or lack of body bending motion in the gait. There were thirty-eight total gaits, each of which was tested three times. The trials measured the distance traveled per gait cycle, which was averaged over every set of trials. The presence of body bending motion created a significant improvement in distance traveled per gait cycle, though it followed a similar trend to the trials without body bending where some gaits were more effective than others. Additionally, the alternating tripod gait was not the most effective gait for forward progress. These results indicate that body bending motion utilized correctly is valuable for efficient locomotion. Also, the robophysical model shows that altering these gait parameters can significantly impact the effectiveness of the gait. These trials have led to a better understanding of hexapod locomotion.

Biological Movement Using Soft Electromagnetic Actuators
Yaw Mensah, University of Tennessee—Knoxville
Faculty Advisor: Ellen Yi Chen Mazumdar, School of Mechanical Engineering

Soft materials and soft actuation concepts have generated new design and control approaches in areas from robotics to wearable devices. However, soft robots currently require the use of rigid materials for externals pumps, valves, and motors for actuation. These rigid materials decrease the portability and versatility of soft robotic systems, preventing them from reaching their true potential. Soft electromagnetic actuator architectures, on the other hand, can potentially solve this problem using the compliant materials for actuator design. These soft electromagnetic actuators can have similar performance to traditional hard linear or rotary electromagnetic motors. In addition, soft materials enable infinite degrees of freedom for bending and morphing. The design of these actuators requires soft conductive materials and soft magnetic materials. In this work, soft permanent magnetic materials are created using silicone and neodymium iron boron composites.  To enable morphing with soft conductors, liquid metals such as gallium and Galinstan (gallium indium tin eutectic) are used. Liquid metals are poured into soft silicone channel molds that are configured to created strong, uniform magnetic fields. In this work we show that the soft coils and soft magnets produced with these processes produced sufficient forces to generate usable motions to mimic biological systems like sea anemones. However, in terms of total displacement, tradeoffs exist between the weight of soft magnets and the magnetic fields they produce.

Autonomous Inter-Stimulus Interval Adjustment for Instantaneous Neuromodulation via Mechanical Stimulation and Paired Brain Stimulation
Heriberto Andres Nieves Vazquez, Florida International University
Faculty Advisor: Jun Ueda, School of Mechanical Engineering

Paired brain stimulation involves applying transcranial magnetic stimulation (TMS) to a motor cortex region of the brain and peripheral stimulation to the corresponding region of the body within a particular inter-stimulus interval (ISI). This paired stimulation induces momentary enhancement of motor function by producing excitability changes in the brain. The objective of this research is to develop an autonomous ISI adjustment procedure for quicker detection of the most effective ISI for an individual undergoing this paradigm. To do so, LabVIEW and Matlab were both used to design a program that would control a previously developed mechanical stimulation (Mstim) robot (i.e., peripheral stimulation) and a TMS to administer the whole procedure automatically. To detect the effectiveness of this therapeutic intervention, a program was used to detect the motor evoked potential (MEP) in electromyography (EMG) recordings during paired brain stimulation. In the future, the ISI along with the MEP would then be entered autonomously into a previously developed program that statistically decides the next testable ISI. Doing so would not only decrease the search time of the effective ISI window of an individual but allow the search to be done in real-time. Ultimately, the program developed was able to run the paired brain stimulation by adjusting the ISI accurately with a standard deviation of 1.93 ms, while also performing offline detection of the MEP.

REACH: Rehabilitation for Augmenting Children Healthcare
Sidney Wise, Georgia Southern University
Faculty Advisor: Ayanna Howard, School of Interactive Computing

The objective of the Rehabilitation for Augmenting Children Healthcare (REACH) project is to investigate the effect of a social robot’s corrective feedback on a user’s reaching kinematics while performing real-world reaching tasks. The project is projected to be assessed by typically developing (TD) adults and children as well as children diagnosed with cerebral palsy (CP). To determine the effectiveness of the feedback provided by the humanoid robot, we had participants play a physical therapy game we developed called Box Top which was designed to improve the player’s reaching kinematics. This paper only delves into the development of Box Top, the design for the experimental procedures, and future plans. Box Top has two major components the electronic system, which is comprised of sensors, LEDs, and all the physical elements of the game, and the software system, which handles data recording, calculation, tester interface, and all the non-physical elements. The game will begin with a practice trial to help the player get an understanding. Then a full game where the human therapist gives feedback to the player. Afterwards player plays by themselves with the goal of trying to hit the targets as fast as possible. Then we let them rest for 2-3 minutes. During that break, the data taken from the previous rounds is used to calculate a baseline for our humanoid robot to use for feedback messages in the final game. There were several bugs with the vision tracking but once their fixed we can move to testing in the future.