Thesis Title: Towards Autonomous Factories: Intelligent Robotic Systems for Task-Oriented Manufacturing via Deep Learning
Embedded Files
Design and implementation of a lance for performing robot-assisted metallurgical sampling and temperature measurement and robot trajectory planning.
Occlusion-resilient pose estimation of textureless components in cluttered environment and its implementation in robotic bin picking.
Development of a three-finger adaptive robotic gripper for handling irregular and complex objects and its dynamic grasp synthesis
A Genetic algorithm based adaptive and hybrid pathfinding algorithm to enhance the performance of robotic bin-picking
Multi-sensor integrated robotic system for hazardous jobs in manufacturing industry
Designed and implemented a robotic system for automatically performing sampling and temperature measurement of molten metal from the furnace.
2. Safe and faster operation as compared to manual sampling including completely removing manual intervention during sampling operation as well as during loading/ unloading the probe from probe.
3. Offline validation of the generated robot program by animated simulation to check for collision and estimation of robot's cycle time. Decrete event simulation has been carried out in KUKA sim software to check the collision detection, joint constraint feasibility and cycle time.
4.Dynamic trajectory planning
5.Control and monitoring of all activities from a central control unit and sensor integration for interlocking.
Occlusion-resilient pose estimation of textureless components in cluttered environment Access Paper
Development of a three-finger adaptive robotic gripper for handling irregular and complex objects and its dynamic grasp synthesis
We introduce underactuated three-finger gripper featuring a novel parallelogram four-bar mechanism with dynamically reconfigurable bases. This low-cost design enables passive adaptability to arbitrary object geometries by autonomously adjusting contact points and force distributions based on an object’s center of gravity (CG) and inertial properties. The reconfigurable base topology permits workspace modulation, allowing the gripper to switch between precision grasps via finger coordination and power grasps via force-optimized enveloping. This capability reduces the risk of slippage and improves the overall success rate of handling complex objects.
We present a physics-based robotic grasping simulation framework for grasp synthesis that systematically models frictional contact interactions, multi-body kinematics, and quasi-static force/torque distributions between reconfigurable grippers and target objects. The framework evaluates grasp stability through slip prediction (via Coulomb friction constraints) and force closure analysis while accounting for the gripper’s reconfigurable joint kinematics and torque limits. This enables virtual design optimization of the three-finger mechanism’s geometry, contact point selection, and actuator torque profiles prior to physical prototyping. The simulator further supports generative exploration of grasp strategies across variable object geometries (prismatic, cylindrical, irregular) and material properties (rigid, compliant), with dynamic adaptation to geometric and inertial uncertainties.
We validated robotic platform trials to check gripper's operational adaptability and robustness, demonstrating 95% success rate across handling various objects, including intricate gears, shafts, plastic objects, 3D printed objects, and ceramic objects,.
Simulation study has been carried out to check the gripper performance.
Deep learning based grasp prediction based on heatmap generation for complex shape geometries (Manuscript under preparation)
Deep learning based grasp prediction based on heatmap generation for complex shape geometries (Manuscript under preparation)
In this work we introduces Grasp detectection CNN network for grasp point detection for complex shape geometries.
With our proposed network we found that our network can detect stable grasp points upto 0.97 IOU and with 98% accuracy.
We also introduces two loss functions MSE and KL divergence with L1 smoothing.
A Genetic Algorithm Based Adaptive and Hybrid Pathfinding Algorithm to Enhance the Performance of Robotic Bin-Picking
The following are the contributions of this paper:
This work introduces an innovative GA-based AHPFA algorithm designed for optimum path planning for bin-picking operations with deep bins and obstructive walls that complicate robotic arm movement for picking. The algorithm is validated using bin picking performed with a vision guided Motoman MH5F industrial robot in both simulation and real-world experimental settings. The proposed AHPFA approach is found to be capable of dynamically adjusting the robot's path ensuring smooth, collision-free movements even with randomly located parts and confined 3D spaces.
Comparative analysis of the GA-based AHPFA algorithm against state-of-the-art metaheuristic algorithms, including ACO, PSO and traditional GA is presented. The proposed GA-based AHPFA is found to be better at handling obstacle avoidance tasks within challenging bin-picking scenarios, significantly reducing computational time and path length.
Page updated
Google Sites
Report abuse