Context:
Traditionally, design of robots and their actuation systems relied heavily in the intuition of its designers. This rarely leads to an optimal robot design in terms of capabilities or (energy) efficiency, with designs chosen based on conservative heuristics and/or over-dimensioned components.
Simultaneously, the need for more dynamic robots has promoted the development of actuators using high torque electrical motors combined with lower gear ratios – quasi-direct drive or proprioceptive actuation. These actuators have reduced reflected inertia, higher speeds and efficiency due to reduced gearing losses, and generally reduced output impedance with better interaction capabilities.
However, the high-torque motors that enable them generate significant heat, and thermals often become the limiting factor in the robot’s performance by reducing available torque as motors reach their temperature limits.
State-of-the-art:
Recent work by us and other authors has led to a number of computational co-design approaches, where the robot’s design and behaviour (control) are simultaneously optimised, typically with the goal of finding the best robot design for a certain (set of) task(s) (and potentially a controller to go with it). We are primarily interested in the design aspect.
We are addressing this topic through several lines of research: Previous students have developed both 1) co-design frameworks based on trajectory optimisation, capable of selecting optimal motor and gearing design parameters for a given robot and including (parallel/biarticulated) compliant elements, and 2) several co-designed prototype robots, aimed at validating the principles. Both of these areas present opportunities for further development.
Critically for this assignment, none of the co-design methods in the state-of-the-art address the thermal dynamics of actuation systems, instead relying on fixed ‘peak torque’ values for motors/actuators.
This assignment:
This assignment focuses on incorporating thermal dynamics in co-design methods for robotic systems. The aim is to be less conservative in robot design, which will improve both the efficiency and capability of the resulting robots. As this is completely new research, this assignment will focus on the modelling, co-design, and simulation aspects, rather than developing new hardware.
In particular, this will require 1) developing accurate enough but computationally efficient models of actuator thermals, including scaling laws for thermal mass and heat dissipation, and 2) usage of these models in co-design, including the cost functions and constraints associated with them.
We have several existing frameworks, developed with Python, Pinocchio, and CasADI, which can serve as a starting point for the co-design implementation.
To find out more, contact Wesley Roozing.
Requirements:
-
Student MSc Robotics or related field;
-
Reasonable understanding of robot and drivetrain dynamics (e.g. MSc course Modelling and Simulation, this can be discussed);
-
Linux and Python experience preferred;
-
git experience preferred.