Lower limb Rehabilitation Robots (LLRRs) assist in therapeutic tasks that involve gait recovery and joint mobility recovery of the lower limbs, in patients recovering from neurologic injuries such as stroke as well as spinal cord injury. LLRRs can sometimes be driven by preprogrammed trajectories or Inverse Kinematics (IK) trajectories, which bring increased computational demand and command supported interaction. This paper proposes an interactive control framework for LLRRs using a hybrid mix of Forward Kinematics (FK) driven movement and an offline Voice Conversational Agent (VCA), based on the Vosk speech recognition engine. The framework proposed is modular in nature that is completely “local”, running offline with no need for the Internet, preserving privacy, and not needing to send uploads cards to cloud processing units. Spoken commands, such as “forward”, “backward”, “rest”, “exercise”, and “stop” are mapped to hip and knee joint angles, which are then driven to FK equations for deriving leg segment positions in an ongoing manner. A hybrid MATLAB-Python implementation is used, where MATLAB is used for simulation and animation, while Python captures the audio input and runs the offline speech recognition component. Recognized transcripts are resolved through fuzzy command matching and are followed by a confidence gated execution to improve tolerance to Automatic Speech Recognition (ASR) variability. Under controlled conditions, command recognition accuracy ranging from 80% to 95%, with end-to-end latencies ranging from 0.89 to 1.32 sec. were seen for the evaluated commands. The performance of feasible offline voice guided interaction and reasonably smooth, anatomically consistent motion transitions, as shown in simulation, provide evidence for the working of the proposed architecture. The main contribution of the work lies in the explicit exposure of ASR offline, fuzzy command matching, applying confidence gated execution, and the use of FK based motion generation, all within a lightweight LLRR oriented framework. This should be enough substrate for future hardware validation, and phase synchronized wearables deployment.
