<p>This paper proposes a stiffness control approach for a 10-degree-of-freedom robotic hand, utilizing gesture recognition based on deep learning applied to electromyographic (EMG) and inertial measurement unit (IMU) signal processing. The proposed system integrates a joint stiffness controller that generates bounded control actions, ensuring precise and stable manipulation. A long short-term memory (LSTM) neural network was trained to recognize three distinct hand gestures–cylindrical, spherical, and pinch grasps–based on EMG and IMU signals acquired from the forearm muscles. These gestures are mapped to specific stiffness levels for each robotic joint, enabling the hand to adapt dynamically to different manipulation tasks. The control strategy is validated through a closed-loop stability analysis, confirming robust and reliable performance under varying conditions. Experimental results demonstrate the effectiveness of the proposed algorithms in both gesture recognition and stiffness control. The robotic hand accurately replicates the intended gestures with appropriate stiffness modulation, achieving seamless interaction with objects of diverse shapes and compliance. This study highlights the potential of combining deep learning with EMG and IMU signal processing to enhance the functionality of robotic hands, paving the way for advanced applications in assistive robotics, prosthetics, and human-robot interaction.</p>