Source-linked AI summary
Quasi-Direct Drive Actuation for a Lightweight Hip Exoskeleton with High Backdrivability and High Bandwidth
Shuangyue Yu, Tzu-Hao Huang, Xiaolong Yang, Chunhai Jiao, Jianfu Yang, Hang Hu, Sainan Zhang, Yue Chen, Jingang Yi, Hao Su
TL;DR
Lower-limb wearable robots need actuators that are lightweight, backdrivable, and high-bandwidth, but conventional and series elastic designs face competing requirements. The paper develops and benchmarks a quasi-direct drive hip exoskeleton using a high-torque-density motor and low-ratio gearing, achieving strong mechanical and control performance while simplifying control. The authors report 17.5 Nm nominal torque, 0.4 Nm backdrive torque, 62.4 Hz bandwidth, and 1.09 Nm RMS tracking error, but no human performance study is included.
Problem
Wearable-robot actuators must combine low mass, high backdrivability, and high bandwidth, while existing designs compromise among these requirements.
Method
The paper develops a portable hip exoskeleton using a high-torque-density motor with 8:1 low-ratio gearing and benchmarks it against conventional and series elastic actuation.
Results
The QDD exoskeleton achieves 17.5 Nm nominal torque, 0.4 Nm backdrive torque, and 62.4 Hz bandwidth, while outperforming benchmarked state-of-the-art devices across reported performance metrics.
Takeaways & Limitations
The results support QDD actuation as a mechanically versatile approach that can provide high performance without complicated control methods.
Takeaways & Limitations
The study quantitatively characterizes actuation and does not include a human performance study of kinetics or electromyography.
Abstract
from arXiv · showhide
High-performance actuators are crucial to enable mechanical versatility of lower-limb wearable robots, which are required to be lightweight, highly backdrivable, and with high bandwidth. State-of-the-art actuators, e.g., series elastic actuators (SEAs), have to compromise bandwidth to improve compliance (i.e., backdrivability). In this paper, we describe the design and human-robot interaction modeling of a portable hip exoskeleton based on our custom quasi-direct drive (QDD) actuation (i.e., a high torque density motor with low ratio gear). We also present a model-based performance benchmark comparison of representative actuators in terms of torque capability, control bandwidth, backdrivability, and force tracking accuracy. This paper aims to corroborate the underlying philosophy of "design for control", namely meticulous robot design can simplify control algorithms while ensuring high performance. Following this idea, we create a lightweight bilateral hip exoskeleton (overall mass is 3.4 kg) to reduce joint loadings during normal activities, including walking and squatting. Experimental results indicate that the exoskeleton is able to produce high nominal torque (17.5 Nm), high backdrivability (0.4 Nm backdrive torque), high bandwidth (62.4 Hz), and high control accuracy (1.09 Nm root mean square tracking error, i.e., 5.4% of the desired peak torque). Its controller is versatile to assist walking at different speeds (0.8-1.4 m/s) and squatting at 2 s cadence. This work demonstrates significant improvement in backdrivability and control bandwidth compared with state-of-the-art exoskeletons powered by the conventional actuation or SEA.
I. INTRODUCTION
Lower-limb wearable robots need actuators that combine low mass, backdrivability, bandwidth, and torque, but conventional and elastic approaches trade these properties off. The paper addresses this gap with a quasi-direct-drive hip exoskeleton and benchmarks its interaction performance.
- I. INTRODUCTION: Conventional actuators have high mechanical impedance, while series elastic actuators improve backdrivability but sacrifice bandwidth, mass, dimensions, or system simplicity.These trade-offs are especially relevant during repeated, high-force human interaction in locomotion.
- I. INTRODUCTION: The bilateral exoskeleton weighs 3.4 kg overall and provides 0.4 Nm backdrive torque, 17.5 Nm nominal torque, and 62.4 Hz control bandwidth.These values are reported as core mechanical performance characteristics of the design.
- I. INTRODUCTION: The paper combines a unified human-robot interaction model with benchmark comparisons across conventional, series-elastic, and quasi-direct-drive actuation.The comparison covers backdrivability and control bandwidth, while the exoskeleton is characterized during walking and squatting assistance.
II. DESIGN REQUIREMENTS
The design requirements prioritize natural hip motion, high torque density, and embodied physical intelligence for wearable assistance. The proposed motor addresses the torque-density limitation of low-ratio transmission through electromagnetic and mechanical optimization.
- II. DESIGN REQUIREMENTS: The exoskeleton accommodates sagittal flexion/extension and frontal-plane abduction/adduction, with a range of motion larger than standard walking requirements.The expanded range is intended to handle a heterogeneous population and varied activities.
- II. DESIGN REQUIREMENTS: Quasi-direct drive uses a high-torque-density motor with low-ratio transmission to provide high backdrivability and bandwidth without relying on active control.Its low mechanical inertia and simplified structure are part of the intended physical intelligence.
- II. DESIGN REQUIREMENTS: Low gear ratios limit output torque density, so the custom actuator combines high nominal torque, reduced speed, and a high torque-inertia ratio.The design specifically targets the principal torque-density barrier of quasi-direct drive.
- II. DESIGN REQUIREMENTS: The custom motor produces about 2.165 Nm nominal torque with 8 Nm/kg torque density and a 2.2 Nm/kg·cm2 torque-inertia ratio.Compared with the Maxon EC flat 505592 benchmark, the design achieves about 11 times higher torque density and 16 times higher torque-inertia ratio.
B. Quasi-direct Drive Actuator
The hip exoskeleton integrates compact actuation, sensing, control electronics, and compliant attachment features to support natural movement. Its mechanical layout provides one active sagittal degree of freedom while preserving passive motion in other planes.
- B. Quasi-direct Drive Actuator: The integrated actuator combines a high-torque-density motor, 8:1 planetary gearbox, magnetic encoder, and controller in a 777 g package.The integrated design is intended to reduce mass and dimensions while supporting safe operation.
- B. Quasi-direct Drive Actuator: The exoskeleton uses one active sagittal degree of freedom and passive frontal and transverse degrees of freedom for natural hip motion.Hinge joints provide abduction/adduction, while elastic thigh straps permit internal/external rotation.
- B. Quasi-direct Drive Actuator: The mechanical system consists mainly of a waist frame, two actuators, two torque sensors, and two thigh braces.The waist frame conforms to the pelvis, and its wide belt increases contact area while reducing pressure.
- B. Quasi-direct Drive Actuator: The robot is designed for large-range activities including stair climbing, sit-to-stand, and squatting.Its passive joints and attachment structure support these activities while allowing hip abduction and adduction.
- B. Quasi-direct Drive Actuator: The electrical system supports torque control, motor control, sensing, communication, and power management through local motor controllers and a high-level microcontroller.The local controller handles current, velocity, and position control, while the high-level controller performs torque control.
B. Human-Exoskeleton Coupled Dynamic Model
The model represents the exoskeleton and human leg as coupled motor, gearbox, transmission, wearable-structure, and human-limb dynamics. It relates voltage and human motion to assistive torque while analyzing bandwidth and backdrivability through gear ratio and transmission properties.
- Model structure: The coupled model comprises motor, speed-reduction, transmission, wearable-structure, and human-leg modules connected through spring-damper force and motion transmission.The human thigh dynamics include orthosis and limb inertia, while assistive torque may be assistive or resistive.
- Motor and gearbox: Motor torque is generated electrically from input voltage and current, then opposed by rotor inertia, viscous friction, and gearbox input torque.The motor equation uses rotor inertia, friction, and motor angle to characterize the input-side dynamics.
- Motor and gearbox: The gearbox reduces output angle and amplifies torque through a gear ratio n:1 before transmitting torque through stiffness k_c and damping b_c.The transmission may use rigid linkages, springs, or cable-based mechanisms represented by equivalent stiffness and damping.
- Transfer functions: The open-loop transfer model relates voltage V(s) and human hip motion θ_h(s) to human-robot interaction torque τ_a(s).G1 maps voltage input to interaction torque, whereas G2 maps human motion input to interaction torque.
- Bandwidth and backdrivability: The effective inertia J_e = n^2J_m and natural frequency depend on gear ratio n, transmission stiffness k_c, and motor inertia J_m.Smaller gear ratio, smaller reflected motor inertia, and larger transmission stiffness increase the natural frequency and open-loop force-control bandwidth.
- Bandwidth and backdrivability: Smaller gear ratio n, transmission damping b_c, or transmission stiffness k_c reduces resistive torque and improves backdrivability and output-link impedance.The limiting cases show negligible resistive torque as n approaches zero and transmission-dominated resistance as n becomes large.
C. Actuation Paradigm Based Hip Exoskeleton Benchmark
The paper benchmarks conventional, series elastic, and quasi-direct drive actuation using a human-exoskeleton interaction model, emphasizing bandwidth, backdrivability, torque capability, and mass. The QDD design achieves high simulated bandwidth and backdrivability while retaining broader actuator versatility, although model simplifications can cause deviations from actual values.
- Actuation Paradigm Benchmark: The benchmark compares conventional actuation, SEA, and QDD using resistive torque and torque-control bandwidth as key versatility metrics.The model represents the actuation paradigms through differences in transmission stiffness, gear ratio, and motor type.
- Actuation Paradigm Benchmark: 73.3 Hz bandwidth and 0.97 Nm backdrive torque demonstrate the QDD actuator’s simulated combination of high bandwidth and backdrivability.The simulation compares conventional actuation, SEA, and QDD under closed-loop torque-control and resistive-torque conditions.
- Actuation Paradigm Benchmark: The QDD benchmark additionally evaluates torque inertia ratio, bandwidth, backdrivability, and output torque density for portable hip-exoskeleton performance.Table III reports the comparison across these actuator-performance factors.
- Actuation Paradigm Benchmark: The simulation may deviate from actual values because of model simplification and variable parameters such as human limb inertia.The authors state that the analysis nevertheless captures underlying characteristics of the three actuation paradigms.
V. CONTROL STRATEGIES
The controller uses hierarchical gait estimation, torque-profile generation, and current-based torque control to adapt assistance across walking and squatting speeds. Its gait estimator matches offline ground truth closely, supporting speed-robust assistance generation.
- V. CONTROL STRATEGIES: The hierarchical controller combines robust gait-intention detection, assistance-torque generation, and low-level current-based torque control.The architecture uses high-, middle-, and low-level control layers.
- V. CONTROL STRATEGIES: R2 = 0.997 on the test set quantifies the neural-network gait estimator’s performance using thigh-mounted IMU signals.The estimator uses one hidden layer with 30 neurons and is trained on walking and squatting data from three able-bodied subjects.
- V. CONTROL STRATEGIES: The middle-level controller converts estimated gait percentage into desired assistance torque through look-up-table profiles and interpolation.Walking assistance uses a human biological model, while squatting assistance uses a sine wave.
- V. CONTROL STRATEGIES: Fig. 9 shows estimated gait percentage matching offline insole-derived ground truth with R2 = 0.997 despite walking-speed changes.The figure also presents the biological torque generated by the assistance algorithm.
VI. EXPERIMENTAL RESULTS
The experimental program uses five actuator and hip-exoskeleton experiments to characterize the device’s backdrivability and bandwidth. The stated goal is systematic characterization of the exoskeleton’s mechanical versatility.
- VI. EXPERIMENTAL RESULTS: Five experiments characterize the hip exoskeleton’s backdrivability and bandwidth at the actuator and whole-exoskeleton levels.The experiments are intended to systematically characterize the device’s mechanical versatility.
- VI. EXPERIMENTAL RESULTS: The evaluation focuses on demonstrating backdrivability and bandwidth of the hip exoskeleton.These properties are assessed through experiments on both the actuator and the exoskeleton.
- VI. EXPERIMENTAL RESULTS: The experimental section frames mechanical versatility as the central performance target for the characterized device.The stated characterization covers the actuator and hip-exoskeleton system rather than a human performance study.
A. Motor Nominal Current Evaluation
The nominal-current experiment evaluates thermal behavior during continuous stall operation under different output currents. At nominal current, the actuator reaches 17.5 Nm nominal torque while its shell reaches 62.7 ℃ after 15 minutes.
- A. Motor Nominal Current Evaluation: Continuous stall tests under different output currents evaluate how motor winding temperature limits actuator output capability.The experiment measures stator temperature with an embedded sensor and shell temperature using a thermal camera.
- A. Motor Nominal Current Evaluation: 17.5 Nm nominal torque is produced at 7.5 A nominal current, while approximately 42 Nm peak torque is produced at 18 A.Torque values are calculated from the calibrated torque constant.
- A. Motor Nominal Current Evaluation: 62.7 ℃ is the highest actuator-shell temperature after 15 minutes of stall operation at nominal current.The stator and shell temperatures were monitored in a 22 ℃ laboratory environment.
B. Exoskeleton Bandwidth Evaluation
The exoskeleton achieves high torque-control bandwidth across tested torque levels, exceeding the reported SEA benchmark, while hardware effects reduce bandwidth below simulation.
- B. Exoskeleton Bandwidth Evaluation: 62.4 Hz maximum experimental bandwidth demonstrates high-frequency torque control, compared with 5 Hz reported for an SEA exoskeleton.Measured bandwidths were 57.8, 59.3, and 62.4 Hz for 10, 15, and 20 Nm references.
- B. Exoskeleton Bandwidth Evaluation: 62.4 Hz experimental bandwidth is lower than the 73.3 Hz model-based simulation because of lower damping and actuator backlash.
- B. Exoskeleton Bandwidth Evaluation: The actuator surface reached 62.7 ℃ after 15 minutes of continuous 7.5 A operation.
C. Exoskeleton Backdrive Torque Evaluation
Unpowered testing shows that the hip exoskeleton requires only approximately 0.4 Nm of backdrive torque during cyclic hip motion.
- C. Exoskeleton Backdrive Torque Evaluation: Approximately 0.4 Nm maximum backdrive torque indicates low mechanical resistance in unpowered operation.One able-bodied subject moved the hip joint through 32.2° at 1 Hz while wearing the unpowered exoskeleton.
D. Torque Tracking for Walking and Squatting Assistance
The exoskeleton tracks assistance torque during walking at 0.8–1.4 m/s and squatting at 2 s cadence, with an average RMS error of 1.09 Nm.
- D. Torque Tracking for Walking and Squatting Assistance: The proposed actuator provides approximately 20 Nm torque during assistance, exceeding the 12 Nm torque reported for the compared hip device.
- D. Torque Tracking for Walking and Squatting Assistance: 1.15–1.27 Nm walking RMS errors and 0.73 Nm squatting RMS error correspond to 3.65–6.35% of peak torque.Walking speeds were 0.8, 1.1, and 1.4 m/s; squatting used a 2 s cadence.
- D. Torque Tracking for Walking and Squatting Assistance: 1.09 Nm average RMS torque-tracking error across 60 tests corresponds to 5.4% of the maximum desired torque.The tests covered walking and squatting assistance.
- D. Torque Tracking for Walking and Squatting Assistance: The actuator combines 17.5 Nm nominal torque with 0.4 Nm backdrive torque and 62.4 Hz bandwidth without complicated control methods.