Fig. 1 Origami exoshell and experimental setup to evaluate the kinematic estimation of the fusion algorithm: ( a ) bench test setup and ( b ) test with the human user wearing the origami exoshell More

Fig. 2 ( a ) Origami exoshell design; arrows indicate the direction of rotation. ( b ) Hall effect sensors are mounted at each joint, and gyroscopes are mounted within each link. ( c ) The exoshell consists of triangular origami modules that when connected form a serial link robot with discrete jo... More

Fig. 3 Hall effect sensors characterization plot. Two Hall effect sensors are placed on opposite sides of the origami joint, depicted as colored boxes in the illustration. More

Fig. 4 ( a ) Human wearing the origami exoshell robot; torso angles ϕ correspond to the relative orientation between the coordinate frames of the top and bottom body segments. ( b ) Illustration of the origami exoshell robot with state variables; the torso angles ϕ correspond to the relative o... More

Fig. 5 Joint kinematic estimation of individual sensors and the KCKF sensor fusion algorithm. Included are the robot kinematics of ( a ) θ 3 ( x -axis rotation), and ( b ) θ 4 ( y -axis rotation). More

Fig. 6 Comparison of the KCKF and the standard KF. The results include the torso angle: ( a ) in the sagittal plane ( x -axis) and ( b ) in the lateral plane ( y -axis). More

Fig. 7 Drift of kinematic estimation over 1 h More

Fig. 8 Kinematic estimation while wearing the origami exoshell robot. The plots include ( a ) the robot kinematics for θ 2 and ( b ) the human’s torso kinematics ϕ. More

Fig. 1 The DJI M100 (2015) has arms that are 0.31 m long, 13” propellers with pitch of 4.5”, and a total mass of 3.133 kg. The inertial frame (ENU) is fixed on the ground, while the body frame ( x , y , z ) is fixed to the UAV. More

Fig. 2 Inertial frame (world coordinates): ( a ) triangle: ENU position and ( b ) shoelace loop: ENU position More

Fig. 3 Velocity magnitude versus time: ( a ) triangle: velocity magnitude and ( b ) shoelace loop: velocity magnitude More

Fig. 4 Gating network for motor angular velocities More

Fig. 5 Comparing ω 1 2 : the Burgers and Gibiansky models are accurate with some overshoot compared to the telemetry data. The Staples model tends to undershoot the telemetry data even after excessive tuning. Overall, none of the models perfectly match the telemetry data: ( a ) triangle: ... More

Fig. 6 Comparing telemetry and MoE with ±20 confidence band: ( a ) triangle: telemetry versus MoE and ( b ) shoelace loop: telemetry versus MoE More

Fig. 7 The Burgers and Gibiansky models tend to be the most accurate during midflight, while the Staples model tends to be more accurate during takeoff and landing: ( a ) triangle: gating network weights and ( b ) shoelace loop: gating network weights More

Fig. 1 ( a ) Nine oil–water flow patterns (FP) in a horizontal pipe, and ( b ) overall distribution of the FPs in the dataset More

Fig. 2 Flowchart of the repeated train-test split for machine learning algorithms, std: standard deviation More