Intelligent prosthetic hips and knees: actuation, perception, and control

• A Frontiers review covers intelligent prosthetic hips and knees from actuation to perception and control.¹
• Limb prosthetics with perception and control inform sensory feedback and human–machine control; the implementation path is orthotic/robotics without direct neural recording (tier-2 watchlist).^{1 1}

Weekly enrichment (2026-07-20)

• The article is a systematic review, “Intelligent prosthetic hips and knees: from actuation to perception and control,” published in Frontiers in Neuroscience (2025) by Wang, Li, and Yu (Shanghai Engineering Research Center of Assistive Devices).²³
• The review searched IEEE Xplore, PubMed, Web of Science, and SpringerLink for 2010–2024 studies; after two-stage full-text screening by two independent evaluators it identified 162 eligible papers and selected 100 representative studies plus 4 seminal pre-2010 works.²³
• It targets high-level amputation, noting hip disarticulation is the highest lower-limb amputation level and accounts for about 2.2% of all lower-limb amputations, and cites WHO figures of ~1.3 billion people (≈16%) living with significant disability and over 1.7 million lower-limb amputees in China.²
• On actuation, it contrasts passive, active, and emerging hybrid active-passive mechanisms; for example, Ueyama et al. (2020) built a motor-driven hip-disarticulation prosthesis whose bare weight exceeded 11 kg and did not reduce amputees’ energy expenditure, while Li et al. (2025) used a remote-center-of-motion design combining a series-elastic and variable-stiffness actuator for lightweight, compliant assistance.²
• Perception is organized around mechanical, bioelectric (sEMG), biomechanical, and environmental signals, including implant-based magnetomicrometry (tracking magnetic beads in muscle) and non-invasive force myography (FMG).²
• The review flags sEMG’s non-stationarity and variability as barriers, and highlights multimodal fusion—e.g., a dual-modal FMG–IMU system classified with a CNN-BiLSTM model for locomotion-mode recognition and transition prediction.²
• Control strategies are categorized as torque compensation, motion following, and direct intention control, the last including proportional control, neuromuscular Hill-type musculoskeletal models, and machine-learning regression (e.g., Gaussian-process estimation of hip/knee/ankle torque).²
• Identified bottlenecks are signal interference, limited adaptability to dynamic environments, and the absence of effective real-time intention-recognition methods; the authors call for bidirectional hybrid active-passive actuation replicating biarticular muscle energy transfer.²
• BCI/neurotech relevance is tier-2: the pipeline is orthotic/robotic with human-machine-environment intent decoding (EMG/FMG/IMU) rather than direct neural recording, but the intention-recognition and sensory-feedback challenges parallel those in neuroprosthetic control.²⁴

Footnotes

1. https://news.google.com/rss/articles/CBMilAFBVV95cUxQbEN0dHJOck9YcWRzYUNMZ29Hd0ZFakpwSHVONkNXYTVLdnptY2hiamgtTXdDbld1dmRJQzdUSlItUnUtQV9zOUdseF9mQlREQld4U1lrak56ZDVuSjZvMG9TRmdkb3lpM212eG9yaEltT1p6bF92aXlmRTNncnQ1RkpqZTA4eDJzdHlLaDk4bTdRcE1Z?oc=5 ↩ ↩² ↩³

2. https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2025.1690921/full ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

3. https://pmc.ncbi.nlm.nih.gov/articles/PMC12833248/ ↩ ↩²

4. https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2023.1032748/full ↩