Master thesis projects 2026
Dynamic Scene Graph with Human Robot Interaction
Program: MSc. in Robotics
Student: Ualikhan Tukenov
Supervisor: Dr. Almas Shintemirov
Co-Supervisor: Dr. Iliyas Tursynbek
Abstract
This thesis focus on enabling a fixed-base robotic arm to perform language-guided manipulation for extended periods with a human in the loop. A dynamic scene graph is built, comprising all recognized
objects, human hands, and the relations between them, including spatial, semantic, and interaction-based relations, to implement a robot-human object handover task.
Scene Graph-Augmented Soft Prompting for VLA Models: Comparing Textual and Graph Neural Network Encodings for Robotic Manipulation
Program: MSc. in Robotics
Student: Yerdaulet Seidizhappar
Supervisor: Dr. Almas Shintemirov
Co-Supervisor: Dr. Iliyas Tursynbek
Abstract
Vision-Language-Action (VLA) models is a promissing research direction for generalisable robotic manipulation, enabling robots to follow natural language instructions through end-to-end learned policies. However, current VLA models lack explicit spatial and relational reasoning capabilities, processing scenes as raw pixel inputs without understanding structural relationship between objects. This work explores augmenting
the X-VLA model a soft-prompted, flow-matching-based VLA architecture with scene graph representations to improve manipulation performance on a Franka Emika FR3 manipulator.
Error-Aware Path Planning for DXF-Based Robotic Layout Marking
Program: MSc. in Robotics
Student: Amantay Kapyshev
Supervisor: Dr. Almas Shintemirov
Co-Supervisor: Dr. Iliyas Tursynbek
Abstract
This work focus on developing theoretical and software foundations for robotized construction layout marking based on DXF floor plans.
Deep Reinforcement Learning with Dense Reward Shaping and Domain Randomization for Sim-to-Real Transfer in Robotic Push Manipulation
Program: MSc. in Robotics
Student: Dalel Abekenov
Supervisor: Dr. Almas Shintemirov
Co-Supervisor: Dr. Matteo Rubagotti
Abstract
This thesis is focused on investigation of sim-to-real transfer of DRL policies for planar push manipulation, trained using the Soft Actor-Critic approach. The work is organized as a three-condition ablation study, examining individual effects of reward function design and domain randomization, followed by zero-shot deployment of the candidate policy on an experimental robot.
Learning Manipulation Skills for the Franka FR3 Using Behavior Cloning: From Simulation to Real-Robot Deployment
Program: MSc. in Robotics
Student: Aiym Issatayeva
Supervisor: Dr. Almas Shintemirov
Co-Supervisor: Dr. Matteo Rubagotti
Abstract
This thesis presents a behavior cloning pipeline for object manipulation on a Franka FR3 robot, including simulation validation, real-robot data collection, policy training, and the deployment of hardware.
Master thesis projects 2021
Model Predictive Control of an Assistive Robotic Arm for Shared Autonomy Implementation
Program: MSc. in Robotics
Student: Anton Kim
Supervisor: Dr. Almas Shintemirov
Abstract
This work describes MPC implementation for 6-DOF robotic arm which can autonomously reach a target pose generated by a grasping pose estimation algorithm and allows the user to send Cartesian velocity commands to the robot via 3-axis joy-stick thanks to the system model based on Jacobian. The developed controller makes use of variable weighting matrices to make movement in different control modes more smooth and realistic. It also includes a visual constraint to minimize occlusion of target object by the manipulator. The next step of this project is to implement a shared autonomy scheme based on the developed MPC framework.
High Level Robot Motion Planning Using POMDP
Program: MSc. in Computer Science
Student: Shyrailym Shaldambayeva
Supervisor: Dr. Almas Shintemirov
Abstract
Motion planning in uncertain environments is an essential feature of autonomous robots. Partially observable Markov decision processes (POMDP) provides a principled approach for such planning tasks and have been successfully employed for various robotic applications. Offline planning algorithms for POMDPs have proved to achieve optimal policies. However, these algorithms are computationally very expensive andare not often applicable to accomplish realistic robot motion planning scenarios. Asan alternative to offline planning Monte Carlo algorithms for online planning were developed, e.g. DESPOT and POMCP are implemented for public use within a setof open-source POMDP libraries. The aim of this master thesis is to explore applicability of available open-source POMDP libraries to real-time robot motion planning tasks. We adopt these libraries for the POMDP models proposed in the literature and replicate a number of realistic benchmark experiments.
Master thesis projects 2019
Program: MSc. in Robotics
Student: Alikhan Zhilisbayev
Supervisor: Dr. Almas Shintemirov
Abstract
This thesis focuses on development of a low-cost robotic gripper with three reconfigurable modular (detachable) underactuated fingers. The project consists of three main parts: development of Prototype I and its mechanical assembly; improvement to Prototype II with assembly and system integration; control design and software development.
Program: MSc. in Robotics
Student: Roman Kruchinin
Supervisor: Dr. Almas Shintemirov
Abstract
This master thesis present a control framework for simultaneous localization and mapping (SLAM) and motion control of a skid-steering robot. Firstly, an experimental mobile robot platform is presented and its limitations are discussed following up by the modified robot embedded control system design. Then, ROS based navigation stack is applied for robot simultaneous localization and mapping (SLAM) and global path generation to a destination point. Then, the control framework for global path following for a skid-steering robot is presented based on the nonlinear model predictive control methodology.The framework is tested in the Webots open-source robotics simulation software.
Recognition of 3D Objects for a Robot Arm Using Deep Learning
Program: MSc. in Computer Science
Student: Sergey Soltan
Main Supervisor: Dr. M. Fatih Demirchi (Dept. of Computer Science)
Co-Supervisor: Dr. Almas Shintemirov
Abstract
Accurate object classification and position estimation is a crucial part for executing autonomous pick-and-place operations by a robot and can be realized using RGB-D sensors becoming increasingly available for use in industrial applications. In this master thesis project we present a novel unified framework for object detection and classification using a combination of point cloud processing and deep learning techniques. The proposed model uses two streams that recognize objects on RGB and depth data separately and combines the two in later stages to classify objects. Experimental evaluation of the proposed model including classification accuracy comparison with previous works demonstrates its effectiveness and efficiency, making the model suitable for real-time applications. The experiments performed on the publicly available Washington RGB-D object dataset show that the proposed framework has 98% fewer parameters compared to the state-of-the-art multimodel neural networks with the cost of approximately 5% drop in accuracy.
Master thesis projects 2017
Intuitive Teleoperation of 6-DoF Universal Robots Manipulators in Constrained Workspace using Nonlinear Model Predictive Control
Program: MSc. in Robotics
Student: Tasbolat Taynyazov
Supervisor: Dr. Almas Shintemirov
Co-Supervisor: Dr. Matteo Rubagotti
Abstract
This master thesis project present a novel control framework for intuitive teleoperation of the Universal Robots manipulators in a constrained environment using human upper limb tracking system. At first, the novel hardware and software designs of a 7-DOF wireless human upper limb tracking system are developed. The tracking system consists of separate individual sensor units for easy mounting on human arms or replacement. The presented experimental tests demonstrate the accuracy of the developed system that is achieved via careful sensor calibration and data filtering. Secondly, the Nonlinear Model Predictive Control (NMPC) technique is used to formulate the UR manipulators joint velocity control loop in order to integrate the human upper limb tracking system for smooth and accurate intuitive robot teleoperation through human operator’s motion tracking. The limitations of the UR manipulators are taken into account as the NMPC constraints while the UR forward kinematics problem is used in the optimization function for predictive control of the robot. The performance of the proposed UR manipulators teleoperation framework was experimentally demonstrated on a real UR5 manipulator equipped with a Robotiq 3-finger adaptive gripper. The proposed approach demonstrated superior performance in completing pick-and-place experimental test operations compared to the same task execution using a standard UR teach peadant.
Mechanical Design and Kinematic Analysis of a Spherical Parallel Manipulator with Coaxial Input Shafts
Program: MSc. in Robotics
Student: Iliyas Tursynbek
Supervisor: Dr. Almas Shintemirov
Abstract
In this thesis a spherical parallel manipulator with coaxial input shafts (Coaxial SPM) is under study. It is a part of a bigger family of spherical parallel manipulators (SPM) with a special feature of unlimited roll rotation around its axis. This feature makes the Coaxial SPM of high interest for applications in motion control system. First, an approach for obtaining unique forward and inverse kinematics solutions is introduced, in order to relate the angular position of the manipulator servomotors to the position and orientation of Coaxial SPM mobile platform and vice versa. Then, a configuration space of the manipulator is defined by using a numerical procedure, in order to guarantee the absence of singularities and of collision between the manipulator links during the manipulator motion. Afterwards, the Cartesian space of the manipulator is generated. Results of these analyses are applied to the assembled mechanical prototype of Coaxial SPM for experimental verifications.
Real-Time predictive Control of UR5 Robotic Arm Through Human Upper Limb Motion Tracking
Program: MSc. in Mechanical Engineering
Student: Bukeikhan Omarali
Supervisor: Dr. Almas Shintemirov
Co-Supervisor: Dr. Hazrat Ali (Dept. of Mechanical Engineering)
Abstract
This thesis reports the authors’ results on developing a real-time predictive control system for an Universal Robot UR5 robotic arm through human motion capture with a visualization environment built in the Blender Game Engine. The UR5 is a 6 degree of freedom serial manipulator commonly used in academia and light industry. It is a very safe robot by design that comes at a cost of a rather limited API with very little support of real-time operation. The motion tracking is performed by a wireless low-cost inertial motion capture setup produced in-house. The external controller incorporates an iTaSC SDLS IK solver and a Python wrapped C explicit model predictive controller generated using the Multi Parametric Toolbox. The visualisation provides the user with the feedback on the robot’s progress towards the target. It is planned to extend the visualisation to virtual reality in future. Tests have shown that the robot follows the operator’s wrist position and orientation with an average of 0.05sec. time lag in the case when the operator moves under the robot’s velocity and acceleration limits. When the operator moves too fast for the robot to keep up in real-time, the robot is able to catch up with the operator with little or no overshooting. Thesis results are described in a late-breaking report and demo accepted by the 12th annual IEEE/ACM international conference Human-Robot Interaction (HRI2017).