Data-Driven Modeling of Industrial Robot Repeatability Using Ensemble Artificial Neural Networks Under Varying Operational Conditions

A systematic AI-based approach for inverse dynamics modelling and validation on UR5 manipulator

Robotic analysis uses mathematical modeling and digital techniques to represent and study mechanical, electrical, and computational systems. Combining these models with physical robots enhances their functions, and artificial intelligence adds intelligence, enabling robots to perform more complex tasks. Inverse dynamics is a key component, mapping joint positions, velocities, and accelerations to the torques needed for accurate motion. This study presents a unified framework that integrates inverse dynamics learning, adaptive control, multi objective optimization, stability verification, and sensorless torque estimation for a 6 DOF UR5 manipulator. A structured dataset representing robot motion with controlled noise and uncertainty conditions is used to train three learning models: Deep Neural Networks, Type 2 Fuzzy Systems, and Gaussian Process Regression. The results show that Gaussian Process Regression achieves the highest torque prediction accuracy, with RMSE of 7.87 Nm and R2 of 0.892, closely followed by the Deep Neural Network (RMSE 11.90 Nm, R2 0.753). Integration with an adaptive controller reduces the trajectory tracking error to 0.0473 rad, which is further reduced to 0.0238 rad using Particle Swarm Optimization, alongside a 5.41% reduction in control energy. Stability analysis using Lyapunov and frequency domain methods confirms bounded, stable closed-loop tracking behaviour, while sensorless torque estimation improves prediction accuracy by 70%. The proposed framework provides a clear methodological foundation for learning-based inverse dynamics modelling and its integration with robotic control systems, and future work will extend the framework to real-time implementation on physical robotic systems.

Sensorless Payload Estimation in an Industrial Robot Using Internal Motor-Current Signals

This article presents a sensorless method for cycle-level payload mass estimation in industrial robots using internal motor-current signals only. The approach is based on automated KUKA Trace acquisition of six-axis current traces from a KUKA KR3 controller, followed by statistical feature extraction and regression using a multilayer perceptron (MLP). Experiments were conducted on two nominally identical KUKA KR3 R540 manipulators under a repeatable handling motion with payloads ranging from approximately 0.4 kg to 2.6 kg, enabling the systematic engineering validation of controller-based current acquisition, feature-based payload regression, and cross-robot validation. This study investigated whether motor-current signals could serve as a reliable virtual sensing source without external force or weight sensors. Under repeated operating conditions, the most accurate MLP configuration achieved a testing mean absolute error (MAE) of 6.75 g and a mean squared error (MSE) of 68.28 g2. With mixed-source training, using data from both robots, the testing MAE decreased to 5.37 g, and the accuracy within a ±15 g tolerance reached 96.88%. In contrast, direct transfer to another nominally identical robot increased the MAE to 80.54 g, revealing a clear cross-robot generalization gap. Overall, this study demonstrates the feasibility of motor-current-based payload verification during repeated single-axis motion under controlled conditions and indicates its potential for gripping validation and missing-part detection in industrial handling applications.

Bootstrapping Self-Supervised Learning of Binary Classification Using Error Bounds: A Case Study on a Robotic Insertion Task

A data engine which gathers data and improves its performance while executing the task, and demonstrates the ability to learn and reduce the need for expensive verifications over time, while staying within the set error-rate.

Integration and Continual Learning-Based Modeling of a Soft Robotic Sensor for Social Robot Proprioception

Generalized Process Simulation of Heterogeneous Industrial Robots Using a Shared Machine Learning Model

Industrial robot process simulation plays an important role in production planning, optimization, and digital twin development. However, many existing simulation approaches rely on robot-specific and manually parametrized models, which limits scalability in heterogeneous robot environments. This paper proposes a unified machine learning–based simulation framework for modelling and simulating process behaviour across different industrial robot platforms using a shared task-space learning policy. The proposed approach combines reinforcement learning and inverse kinematics with clearly separated roles. The reinforcement learning policy generates end-effector motion in task space, while inverse kinematics converts the desired motion into robot-specific joint configurations and supports geometric feasibility. This separation reduces the dependence of the learning model on the joint structure of a particular robot. Experimental evaluation on heterogeneous robot platforms showed that the proposed framework can learn accurate target-reaching behaviour in simulation. The best-performing policy achieved a success rate close to 1.0, a mean final distance error of 0.0082 m, and a mean final orientation error of 5.55°. These results indicate that a shared task-space learning policy combined with robot-specific inverse kinematics can support scalable robot process simulation.

Neural-Augmented Torque Control for Robotic Manipulators: Modeling, Learning, and Real-Time Performance

This thesis investigates advanced modeling and control strategies for robotic manipulators, focusing on the DLR-HIT II robotic hand and the KUKA LBR iiwa. It presents three core contributions that integrate simulation, model-based control, and data-driven methods to improve torque and position control under uncertainties and disturbances. First, a dual-platform simulation framework is developed using MATLAB Simscape Multibody and CoppeliaSim. The system accurately models the DLR-HIT II hand’s kinematics and dynamics, enabling both control validation and realistic interaction with virtual environments. The use of Unified Robot Description Format (URDF)-based modeling sup-ports reusability and modular analysis. Second, a physics-informed neural network (PINN) is proposed for direct torque and position control. This method uses only time and joint position inputs, internally computes derivatives, and generalizes well across various trajectory types. It eliminates the need for separate feedback controllers and shows strong robustness under disturbances, while maintaining low computational cost. Third, a Spike-Aware Hybrid Torque Control (SA-HTC) architecture is enhanced with a feedforward neural network (NN) trained offline. The network learns to improve Computed Torque Control (CTC) for unmodeled effects, friction, and external forces by refining the CTC torque output in real time. Simulation results across diverse trajectories and noise levels demonstrate that the SA-HTC method significantly improves tracking accuracy and robustness compared to classical CTC. To support deployment on position-controlled hardware, a physics-informed torque-to-position interface is introduced and compared with virtual stiffness and admittance wrappers, yielding lower steady-state bias, reduced phase lag, lower tracking error, and robust cross-trajectory generalization. Thus, these contributions advance the integration of learning-based and model-based control strategies in robotics. The results highlight scalable and efficient methods for accurate trajectory tracking and torque control, offering practical potential for robotic manipulation and automation applications.