The system integrates computer vision and RoboDK simulation, enabling real-time detection of geometric shapes such as cylinders, rectangles, and hexagons for precise robotic grasping and placement.

A voice-guided control interface allows users to issue natural spoken commands (e.g., “pick cylinder” or “drop rectangle”) using speech recognition for intuitive human–robot interaction.

The captured voice command is processed through a Large Language Model (LLM) via the Hugging Face API, which classifies and interprets the user’s intent into actionable robotic tasks.

The robot then executes motion planning and control in RoboDK, including pick-and-place, rotation, and shaking operations, using analytical pose transformations and inverse kinematics.

This hybrid framework combines LLM-based understanding with analytical robotics, achieving an intelligent, adaptive, and voice-interactive robotic system for real-time object manipulation.

We have developed a low-cost, deep learning-based warehouse automation system with 100% success rate in real-world applications.

✅ Palletizing Operation – A robotic arm efficiently picks and places bearing housings from a conveyor belt in grid structure of 2*2.

✅ Smart Sorting – A two-finger gripper adapts grasp strategies to sort different objects from a conveyor belt.

Our sensor-integrated deep learning approach enhances production rate, improves efficiency, and reduces operational costs—making automation more accessible for industries. 

We developed an adaptive in-pipe inspection robot designed to autonomously navigate pipelines of varying diameters and complex geometries. The robot is equipped with three spring-loaded legs placed 120° apart around the central body, providing symmetrical support and self-centering capability. This configuration enables the robot to maintain stable motion and adjust its shape dynamically in response to changes in pipe diameter and curvature, allowing it to traverse bends, junctions, and vertical sections efficiently.

For inspection, the robot integrates an onboard vision system coupled with a deep learning model to analyze the internal condition of the pipe in real time. The system is capable of detecting clogging, corrosion, leakage, and surface irregularities with high accuracy. Captured visual data are processed using a convolutional neural network (CNN)–based architecture trained on annotated pipe defect datasets. The inspection data, including live video streams, are transmitted wirelessly to a remote monitoring server for visualization, logging, and further analysis.

This smart inspection platform provides a robust and cost-effective solution for predictive maintenance in industrial pipelines, reducing manual intervention and enabling early fault detection to minimize downtime and operational risks.

In this setup, two ultrasonic sensors are employed to detect objects moving on a conveyor belt in real time. The sensors continuously measure the distance and position of the objects, enabling accurate detection of their presence and motion. This sensor data is then transmitted to the robot controller, which synchronizes the robot’s motion with the conveyor speed to perform precise pick operations. By integrating ultrasonic sensing with robotic control, the system enables efficient and adaptive object picking from a continuously moving conveyor, reducing cycle time and improving automation efficiency.