A narrative review published in Discover Robotics synthesizes recent technological developments in upper limb prosthetics, highlighting a significant shift from traditional myoelectric control to autonomous sensor-fused architectures. The study, authored by Nadia Hachoumi, Rquia Laabidi, and Mohamed Eddabbah, examines how artificial intelligence is transforming prosthetic devices into intuitive bionic extensions. By analyzing peer-reviewed literature from 2013 to 2024, the review identifies a technical revolution in integrated control and functionality, aiming to map the path toward the technological frontier of bionic rehabilitation. The research focuses on the incorporation of machine learning, edge computing, and energy management strategies to enhance the usability and performance of prosthetic limbs for amputees.

The methodology involved a systematic search of PubMed, IEEE Xplore, Scopus, and Google Scholar, yielding peer-reviewed publications within the specified timeframe. Following the PRISMA 2020 protocol, the researchers prioritized quality over quantity, selecting thirty major studies based on hardware validation and clinical outcome reporting criteria. This selective approach ensured that the review focused on studies where AI algorithms were applied to physical prosthetic prototypes or commercial devices, such as the Hero Arm, Taska Hand, or i-Limb. The selection process excluded theoretical simulations or purely computational models, aiming to provide a reliable overview of technologies transitioning from laboratory settings to real-world rehabilitation.

One of the primary findings of the review is the substantial improvement in control success rates achieved through AI-driven architectures. Comparative analyses indicate that Pattern Recognition and Regression-based control systems have improved success rates from approximately 75% in legacy systems to more than 92%. This advancement is attributed to the use of stochastic models, such as Linear Discriminant Analysis and Support Vector Machines, which calculate the likelihood of movement occurrence rather than relying on simple threshold triggers.

Latency remains a critical factor in the usability of prosthetic devices, and the review highlights that state-of-the-art devices now achieve end-to-end control latencies below the human perception threshold of 125 milliseconds. The integration of computer vision into these open control loops further reduces the user’s cognitive load and compensatory shoulder movements by 30%.

Energy management presents another significant challenge in the development of AI-powered prosthetics, as advanced neural architectures inherently increase computational overhead and power consumption. The review notes that emerging energy-harvesting strategies, specifically kinetic energy harvesting from natural limb oscillations, exhibit peak power outputs of 25.8 milliwatts. While this technology suggests an increase in battery longevity by 15.7%, it remains a critical area for future research to ensure that wearable devices maintain thermal stability and sufficient power for continuous operation. The balance between processing power and energy efficiency is essential for the long-term viability of these intelligent bionic systems in daily use.

The evolution of degrees of freedom in prosthetic hands is also a key theme in the review. While a human hand possesses 27 degrees of freedom, replicating this anatomical complexity in wearable technology poses serious engineering challenges due to weight and power constraints. Current commercial benchmarks, such as the i-Limb Quantum and Michelangelo Hand, offer 5 to 6 motorized degrees of freedom, facilitating independent finger articulation and thumb rotation. Research frontiers are exploring prototypes with up to 22 degrees of freedom to achieve near-complete anatomical replication, while devices like the Taska Hand utilize internal degrees of freedom to define 23 different grips, ranging from fine precision pinches to high-torque tool grips.

Shared control architectures represent a frontier in prosthetic design, where AI manages low-level tasks such as holding forces during a grip, while the user provides high-level intent, such as reaching for a target. This division of labor reduces long-term mental fatigue by automating the maintenance of precision grips during prolonged activities of daily living. The integration of reflexive control loops further enhances safety, allowing the AI to increase grip force in response to sensor detection of an object slip, with response times shorter than human neural reaction. This prevents accidental drops and contributes to a more seamless and intuitive user experience.

Despite these advancements, the review acknowledges that a sensory gap persists in current prosthetic technologies. The establishment of haptic feedback through vibration and electrical stimulation is identified as essential for restoring the sense of pressure and texture, thereby providing users with a greater sense of device ownership. Real-time information about hand aperture and wrist rotation has been shown to decrease compensatory movements in the shoulder and torso by 30%, reducing the risk of musculoskeletal disorders and overuse injuries in the sound limb. This bidirectional flow of information transforms the prosthesis from an external tool into a mental extension of the user’s body.

The review also addresses the economic and accessibility barriers that currently limit the global availability of high-performance prosthetic systems. Despite their technical sophistication, these devices remain high-cost mechanical gadgets at the end-user level. The authors suggest that future research must focus on integrating soft robotics, 3D printable customization, and non-invasive neural interfaces to make these systems more accessible and sustainable. By leveraging 3D printing and modular designs, manufacturers can reduce production costs, while public-private partnerships can support the deployment of these technologies for underserved populations worldwide.

Future perspectives outlined in the study emphasize the integration of soft materials and advanced AI to mimic biological tissue and enable lightweight designs. The use of flexible polymers and Shape Memory Alloys in soft robotics, combined with non-invasive brain-machine interfaces for precise control, represents a promising direction for next-generation prosthetics. Additionally, real-time adaptations through Online Reinforcement Learning are expected to enhance responsiveness and address issues such as muscle fatigue. These innovations aim to create dynamic, developing environments where AI-powered prosthetics can operate easily in home settings, ultimately improving the quality of life for individuals with limb loss.

The review distinguishes between deterministic and stochastic control paradigms to explain the shift in prosthetic functionality. Traditional systems operate on conditional logic, where crossing a specific voltage level triggers a binary action. While reliable, this approach lacks the nuance required for complex tasks such as typing. In contrast, AI-driven systems employ stochastic models to calculate the likelihood of a movement occurring. This probabilistic approach allows for more human-like motion but introduces a risk of false positives resulting from signal noise or muscle fatigue. The review identifies Information Throughput as a key metric for evaluating these systems, noting that AI-controlled devices score significantly higher than conventional surface EMG systems. Recent studies indicate that AI agents can decode up to six degrees of freedom simultaneously with above 90% success within a 300-millisecond window, a level of performance not achievable by conventional switching-based control systems.

Hardware optimization is critical for maintaining the performance of these intelligent systems within wearable constraints. The review highlights the use of model optimization techniques, such as quantization, which can decrease memory usage by 75%. This reduction is essential for deploying deep learning models on embedded hardware with limited resources. The integration of high-performance microprocessors, including the ARM Cortex-M series and NVIDIA architectures, facilitates this embedded integration. By processing signals on the device, these systems avoid the delays associated with external computing, ensuring that control loop latency remains under the 125-millisecond human perception limit. This technical architecture supports the transition from laboratory prototypes to practical clinical applications, where real-time processing power demands must be balanced with the requirement for non-invasive operation and minimal time delay.

The review also examines the role of natural language processing in enhancing prosthetic accessibility. By integrating voice commands and verbal interactions, the interface becomes more accessible to users with physical limitations. This capability simplifies the control of complex movements, allowing users to operate their prostheses in a more intuitive manner. The bidirectional control architecture described in the study positions AI as a central processing hub that decodes electromyography intent while simultaneously handling haptic feedback from sensors. Such advancements contribute to a sense of embodiment, transforming the prosthesis from a reactive tool into an intuitive extension of the user’s nervous system.

Limitations in current technology remain a focus of the review’s discussion. Despite high classification accuracies, the practical utility of these systems is constrained by energy management challenges. Advanced neural architectures increase computational overhead, which endangers battery life and thermal stability in wearable devices. The review notes that while energy harvesting from natural limb oscillations offers a peak power output of 25.8 milliwatts, this currently extends battery life by only 15.7%. This modest improvement underscores the need for further innovation in energy-efficient neural networks and adaptive actuation. Additionally, the high cost of these devices at the end-user level limits their global availability. The authors emphasize that future research must address these economic and technical barriers to ensure that high-performance bionic systems are sustainable and accessible to a worldwide amputee population.

Subject of Research: Artificial intelligence integration in upper limb prosthetics

Article Title: A narrative review on advances in upper limb prosthetics empowered by artificial intelligence highlights a technical revolution in integrated control and functionality for the user

Article References: Hachoumi, N., Laabidi, R., & Eddabbah, M. (2026). A narrative review on advances in upper limb prosthetics empowered by artificial intelligence highlights a technical revolution in integrated control and functionality for the user. Discover Robotics, 2(1), Article 3. https://doi.org/10.1007/s44430-026-00020-w

Image Credits: AI Generated

DOI: 10.1007/s44430-026-00020-w

Keywords: upper limb prosthetics, artificial intelligence, myoelectric control, edge computing, bionic rehabilitation, energy harvesting, pattern recognition, haptic feedback, narrative, review, advances, upper