Development of a Robust Parameter Estimation Technique for Dynamical Systems Using Causal Curiosity - A Tracked Vehicle Application

July 3, 2024

Parameter estimation is a cornerstone process in both science and engineering, aiming to indirectly measure the parameters of a system by analysing its outputs, typically to use the resulting estimations in models of the system. Traditional parameter estimation methods often struggle due to the need to derive a specific formulation of the estimation process for each model and parameter. This thesis presents the development of a new parameter estimation methodology leveraging the principles of Causal Curiosity. In contrast with traditional parameter estimation techniques, the proposed method uses an optimization algorithm to explore the actions that can be performed which lead to the most distinguishable values of the outputs, regarding different parameter values. Notably, this method can be applied in simulation and real-world scenarios, enabling the estimation of both intrinsic and extrinsic parameters. The proposed method was validated through extensive simulations and experiments with a tracked vehicle, achieving high accuracy in parameter estimations. Simulation experiments showed relative errors as low as 0.25%, while real-world experiments with a small tracked robot demonstrated errors of 0.76% when estimating the distance between the tracks and 0.37% for the wheel radius. Overall, this research highlights the potential of integrating causal analysis into parameter estimation, offering a robust and flexible tool applicable across diverse scientific and engineering fields.

Authors in the community:

• 

  J

  João Júdice Consolado Lameiras Martins

  ist193271

Supervisors of this institution:

• 

  A

  Alexandra Moutinho

  ist13959

mechanical-engineering - Mechanical engineering

por - Portuguese

Instituto Superior Técnico