Thesis Work in Robotics Motion Control i Västerås

Om tjänsten

You will be part of the motion control group at ABB Robotics. We are responsible for a wide range of areas within the robot controller development spanning from modeling, identification, control design and tuning to signal processing for diagnosis and supervision as well as optimization for path planning.

Here are some possible areas for a thesis project within the motion control area. The actual scope of the project will depend on the applicants cv and the priorities within the development teams:

1. Towards on-line optimization and MPC based dynamic re-planning of the trajectory
The project should investigate the real time processing limits of today’s available optimization routines. The thesis should analyze the real time processing of IP or SQP routines to propose a reasonable trade-off between processing time and accuracy of the solution. The final goal is to understand where the actual bottlenecks (which prevent us to run online optimization) are and propose reasonable solutions where possible.

2. Investigation and verification of a potential new “MoveOPT” instruction based of optimal trajectory computed through off-line optimization.
We are now aware that a powerful optimization methods can tell us which performance are possible to get from a specific robot model, given a set of constraints. The student should set up an optimal trajectory planning problem using CASADI (non-linear programming environment) for different robot models. The thesis should compare the optimal performance achieved with respect to what is possible to get today using traditional move instructions (MoveJ MoveL and the nearest MoveC). Tests on the real robots should validate the results. A sensitivity analysis to the various constraints is also of interest to understand which are the most critical. The goal of the thesis is to have a good estimate of what is possible to achieve using a new “MoveOPT” instruction.

3. Supervision of configuration parameters for industrial robots
The objective of the master thesis is to investigate effective methods to verify if the dynamic parameters are correctly configured. Dynamic parameters of interest are for example: arm loads, payload, gravity vector, friction coefficients, etc. The method should be applicable to any robot model (serial or parallel kinematics) and should be preferably able to be performed on-line during the normal robot operations, without use of specific pre-designed movements. Machine learning techniques or constrained non-linear programming methods (CASADI, IP, SQP bases routines, etc…) should be investigated.

4. Estimation of safety zone for a robot as a function of configuration
This work should lead to a simple and conservative approximation of the zone where the robot can move provided that a stop is ordered at a given position, orientation and speed. The approximation should be built using simulations or optimization where the robot and the brake and actuator constraints are included. It is assumed that a simplified model and also simplified constraints can be used in order to show the feasibility of different approaches. The approximation can then be used in a supervision to protect objects and also humans from robot impact. Within the work a complete setup of the concept should be realized, at least in simulation.

5. Combining AI and ML with simulation
Machine learning is a very promising technique in robotics and it can make it possible to solve tasks autonomously by learning from experiments. One drawback of many of the learning techniques is that a lot of data is necessary to be able to learn and validate an ML solution. If a task, for example a grasp or an assembly operation, should be performed in an application, it is necessary to perform an extensive experimental effort. In this master thesis project simulation tools with realistic robot and environment simulation is used in order to accelerate the learning in order to be able to perform a grasp and assembly task, for example of snap type. The solution should be performed in a simulated environment but if time permits the final learning and validation should be performed in real experiments. A major problem in learning from simulation is the uncertainties of the simulation models and how they are different from the real world. This thesis project should approach this and attempt to find strategies for mitigation.

The master thesis project requires good knowledge in automatic control and signal processing as well as experience of the tools Matlab and Simulink.

Extra information

Status

Stängd

Ort

Västerås

Typ av kontrakt

Heltidsjobb (förstajobb)

Typ av jobb

Produktion / Industri

Körkort önskas

Nej

Tillgång till bil önskas

Nej

Personligt brev krävs

Nej