eess.SYMar 4, 2026
SaveSoft Semi-active Back Support Device with Adaptive Force Profiles using Variable-elastic Actuation and Weight Feedback
Organizations: Arizona State University · Carnegie Mellon University
Abstract
Portable active back support devices (BSDs) offer tunable assistance but are often bulky and heavy, limiting their usability. In contrast, passive BSDs are lightweight and compact but lack the ability to adapt their assistance to different back movements. We present a soft, lightweight, and compact BSD that combines a variable-stiffness passive element and an active element (an artificial muscle) in parallel. The device provides tunable assistance through discrete changes in stiffness values and active force levels. We validate the device's tuning capabilities through bench testing and on-body characterization. Further, we use the device's tuning capabilities to provide weight-adaptive object lifting and lowering assistance. We detect the weight handled by the user based on forearm force myography and upper-back inertial measurement unit data. Furthermore, electromyography analyses in five participants performing symmetric object lifting and lowering tasks showed reductions in back extensor activity. Preliminary results in one participant also indicated reduced muscle activity during asymmetric lifting.
Figures & tables
Figure 1: Overview. (a) Our proposed back support device combines a variable-stiffness passive element and an active element (a pneumatic artificial muscle) in parallel. (b) The device implements weight-adaptive assistance using discrete passive and active force profiles. (c) The lightweight and compact soft structure enables assistance during lifting and lowering, while allowing unrestricted mobility during activities such as sitting and walking.
| Device | Active or Passive | Rigid or Soft | Force Generating Element | Weight (kg) | Peak Torque (Nm) | Torque Density (Nm/kg) | Weight Adaptive Profile | EMG Reduction (%) |
|---|---|---|---|---|---|---|---|---|
| [ 50 ] | Passive | Soft | Elastic band | 1.4 | 35 | 25 | No | NA |
| [ 26 ] | Passive | Rigid | Compliant mechanism | 4 | 30 | 7.5 | No | 8 |
| [ 52 ] | Passive | Rigid | Compliant mechanism | 5 | 7.04 | 1.41 | No | NA |
| [ 25 ] | Passive | Soft | Variable stiffness elastic band | 1.2 | 20.4 | 17 | No | NA |
| [ 32 ] | Active | Rigid | Direct drive motor | 8 | 35 | 4.375 | No | 36.3 |
| [ 34 ] | Active | Rigid | Direct drive motor | 6 | N/A | N/A | No | 11.63 |
Table 1: Comparison of existing untethered BSDs.
Figure 2: (a) Device assembly with the VS resistance band (passive), IPAMs (active), and origami muscle (slack tuning). (b) Device and portable pneumatic/electronic units worn by a user.
Figure 3: (a) Design of the VS resistance band with an integrated electroadhesive (EA) clutch. (b) Principle of operation of a single interface of the EA clutch: When 300 V is applied between the positive and negative electrodes, the electro-adhesion at the interface prevents sliding. This reduces the active length of the resistance band and increases stiffness. The EA clutch includes eight such interfaces.
Figure 4: (a) IPAM design: It consists of a silicone tube helically wound with fiber reinforcement and sealed at both ends. Our device uses two such IPAMs in parallel. (b) Working principle of the IPAM: When pressurized, it extends axially; when deflated, it contracts, thereby generating an active force.
Figure 5: (a) Origami muscle assembly for slack tuning. (b) Working principle of the origami muscle: On vacuuming, a zigzag skeleton housed in a sealed fabric enclosure contracts, shortening the device. (c) Working principle of the EA muscle brakes: When 300 V is applied between the positive and negative electrodes, the electro-adhesion at the interface prevents sliding, locking the origami muscle length after slack tuning. This enables compact actuation without constant power draw. The EA muscle brakes include five EA interfaces on either side of the origami muscle.
Figure 6: (a) FMG armband design: Six FSRs mounted on PET substrates measure radial forearm pressure during weight handling. (b) An inextensible band wrapped over the elastic band limits excessive radial expansion of the elastic band when the user handles a weight. (c) FMG armband worn by a human subject.
Figure 7: (a) Pneumatic system schematic. (b) Truth table depicting the pneumatic control logic.
Figure 8: Schematic of the electronic system used to drive the EA clutch, the IPAMs, and the origami muscle assembly.
Figure 9: Analytical model of the VS resistance band: Spring-network model used to derive stiffness of the resistance band as a function of geometric and material parameters.
Figure 10: Second-order polynomial surface fit showing IPAM force as a function of elongation and pressure, obtained via bench testing.
Figure 11: Analytical model of the full device representing total device force as the sum of passive force from the VS resistance band and active force from the IPAM. The origami muscle assembly is in series but is assumed to be inextensible.
Figure 12: Characterization results. (a) Force profile tuning in the passive mode. (b) Rapid response of the EA clutch. (c) Force profile tuning in the active mode. (d) Peak device force with combined EA clutch and IPAM actuation. (e) Rapid response of the IPAM and the accurate prediction of the device force by the analytical model. (f) Dynamic force testing in the active mode and the accurate prediction of the dynamic device force by the analytical model. The shaded region represents standard deviation in all the sub-figures.
Figure 13: Device force profile optimization for object lowering and lifting. (a) Coupled device-musculoskeletal model. (b) Optimal force profile for object lowering. (c) Optimal force profile for object lifting.
Figure 14: State and weight estimation. (a) Overview of the automated weight adaptive force profile tuning algorithm. (b) Labeling of the training data for the state and the weight classifier. (c) Real-time state and weight classification in a representative subject. (d) Real-time state and weight classification accuracy across three subjects.
Figure 15: On-body Device Characterization. (a) Experimental protocol: Motion capture used to determine the trunk angle and the RoM, load cell used to measure the device force, and the wire draw encoder used to measure device elongation. (b) Force profile tuning in the passive mode that facilitates weight-adaptive lowering assistance. (c) Force profile tuning in the active mode that facilitates weight-adaptive lifting assistance. (d) Force vs device elongation in the active mode. (e) Trunk angle vs device elongation in the active mode. (f) Increasing energy added to the human body over a flexion-extension cycle with the increasing active force level.
Figure 16: (a) Experimental setup and protocol for (a) the symmetric lifting/lowering task. (b) the asymmetric lifting/lowering task.
Figure 17: Evaluation of back extensor and trunk flexor activity during symmetric object lifting and lowering. (a) Effect on the back extensors during weightlifting: The peak normalized root mean squared (RMS) EMG of ESL (mean of left and right) for individual subjects during weightlifting without and with the device. (b) Effect on the trunk flexors during weightlifting: The peak normalized RMS EMG of RA (mean of left and right) for individual subjects during weightlifting without and with the device. (c) Effect on the back extensors during weight lowering: The peak normalized RMS EMG of ESL (mean of left and right) for individual subjects during weight lowering without and with the device. (d) Effect on the trunk flexors during weight lowering: The peak normalized RMS EMG of RA (mean of left and right) for individual subjects during weight lowering without and with the device.
Figure 18: Evaluation of back extensor and trunk flexor activity during asymmetric object lifting and lowering (a) Reduction in the peak normalized RMS EMG of left ESL, right ESL, left EO, and right EO during asymmetric weight lifting in a single subject with the device. (b) Reduction in the peak normalized RMS EMG of left ESL, right ESL, left EO, and right EO during asymmetric weight lowering in a single subject with the device.
More in this topic
Assistive Robotics
Assistive robotics develops robots and wearable devices that support mobility, daily activities, rehabilitation, and care.
See all 207 papers