[Theory] REMINDER: 8/17 Thesis Defense: Chip Schaff, TTIC

[Mary Marre]

*Thesis Defense: Chip Schaff, TTIC*

*When:  *     Wednesday*,* August 17th at *9:00 - 11:00 am CT*


*Where:       *Talk will be given *live, in-person* at

                   TTIC, 6045 S. Kenwood Avenue

                   5th Floor, Room 530



*Virtually: *  *attend virtually here
<https://uchicagogroup.zoom.us/meeting/register/tJIrf-2upjwjE9TDxN4DCKHjYU2Y18wUEMui>*



*Who: *        Chip Schaff, TTIC


*Thesis Title:* Neural Approaches to Co-Optimization in Robotics

*Abstract: *Robots and intelligent systems that sense or interact with the
world are increasingly being used to automate a wide array of tasks. The
ability of these systems to complete these tasks depends on a large range
of technologies such as the mechanical and electrical parts that make up
the physical body of the robot and its sensors, perception algorithms to
perceive the environment, and planning and control algorithms to produce
meaningful actions. These components have strong dependencies between them.
For example, robots will perform better when their bodies admit dynamics
that are well suited for the control problems that they regularly
encounter, and perception systems perform better with appropriate sensor
design and placement. Therefore, it is often necessary to consider the
interactions between

these components when designing an embodied system.



This thesis explores work on the task-driven optimizing of robotics systems
in an end-to-end manner, simultaneously optimizing the physical components
of the system with inference or control algorithms for task performance.
Through the study of specific problems, such as beacon-based localization
and legged locomotion, we develop a learning-based framework to co-optimize
all aspects of robotics systems. In this way, this thesis makes strides
towards an efficient and automated approach to the design of robotics
systems tailored to a specific application, which has the potential to both
improve the performance of robotics systems and reduce the cost and barrier
to entry of robot design.



We start by considering the problem of optimizing a beacon-based
localization system directly for localization accuracy. Beacon-based
localization is a popular approach in environments where GPS is
unavailable, such as underwater, underground, or indoors. Designing such a
system involves placing beacons throughout the environment and inferring
location from sensor readings. The space of algorithms to automatically
design these systems is relatively unexplored and past work often optimizes
placement and inference separately. In our work, we develop a deep learning
approach to optimize both beacon placement and location inference directly
for localization accuracy. In simulated experiments, our approach
significantly outperforms strategies that consider beacon placement and
location inference

separately.



We then turn our attention to the related problem of task-driven
optimization of robots and their controllers. Approaches that automate the
design of robots have a long history and include several techniques such as
evolutionary algorithms, trajectory optimization, and nonlinear
programming. Reinforcement learning has proven successful at solving
complex control problems but, at the start of our work, it was largely
unexplored for co-optimization. Therefore, we start by proposing a
data-efficient algorithm based on multi-task reinforcement learning. Our
approach efficiently optimizes both physical design and control parameters
directly for task performance by leveraging a design-conditioned
controller capable
of generalizing over the space of physical designs. We then follow this up
with an extension to allow for the optimization over discrete morphological
parameters such as the number and configuration of limbs. Finally, we
conclude by exploring the fabrication and deployment of optimized robots.
In this work we extend our previous algorithm to allow for the
co-optimization of soft crawling robots, develop techniques for
speeding up finite
element simulations, and successfully fabricate and transfer the optimized
robot from simulation to the real world.

*Thesis Committee:*
*Matthew R. Walter <mwalter at ttic.edu>* *(Thesis Advisor)*
Ayan Chakrabarti
Audrey Sedal
David McAllester


Mary C. Marre
Faculty Administrative Support
*Toyota Technological Institute*
*6045 S. Kenwood Avenue*
*Chicago, IL  60637*
*mmarre at ttic.edu <mmarre at ttic.edu>*


On Thu, Aug 11, 2022 at 2:41 PM Mary Marre <mmarre at ttic.edu> wrote:

> *Thesis Defense: Chip Schaff, TTIC*
>
> *When:  *     Wednesday*,* August 17th at *9:00 - 11:00 am CT*
>
>
> *Where:       *Talk will be given *live, in-person* at
>
>                    TTIC, 6045 S. Kenwood Avenue
>
>                    5th Floor, Room 530
>
>
>
> *Virtually: *  *attend virtually here
> <https://uchicagogroup.zoom.us/meeting/register/tJIrf-2upjwjE9TDxN4DCKHjYU2Y18wUEMui>*
>
>
>
> *Who: *        Chip Schaff, TTIC
>
>
> *Thesis Title:* Neural Approaches to Co-Optimization in Robotics
>
> *Abstract: *Robots and intelligent systems that sense or interact with
> the world are increasingly being used to automate a wide array of tasks.
> The ability of these systems to complete these tasks depends on a large
> range of technologies such as the mechanical and electrical parts that make
> up the physical body of the robot and its sensors, perception algorithms to
> perceive the environment, and planning and control algorithms to produce
> meaningful actions. These components have strong dependencies between them.
> For example, robots will perform better when their bodies admit dynamics
> that are well suited for the control problems that they regularly
> encounter, and perception systems perform better with appropriate sensor
> design and placement. Therefore, it is often necessary to consider the
> interactions between
>
> these components when designing an embodied system.
>
>
>
> This thesis explores work on the task-driven optimizing of robotics
> systems in an end-to-end manner, simultaneously optimizing the physical
> components of the system with inference or control algorithms for task
> performance. Through the study of specific problems, such as beacon-based
> localization and legged locomotion, we develop a learning-based framework
> to co-optimize all aspects of robotics systems. In this way, this thesis
> makes strides towards an efficient and automated approach to the design of
> robotics systems tailored to a specific application, which has the
> potential to both improve the performance of robotics systems and reduce
> the cost and barrier to entry of robot design.
>
>
>
> We start by considering the problem of optimizing a beacon-based
> localization system directly for localization accuracy. Beacon-based
> localization is a popular approach in environments where GPS is
> unavailable, such as underwater, underground, or indoors. Designing such a
> system involves placing beacons throughout the environment and inferring
> location from sensor readings. The space of algorithms to automatically
> design these systems is relatively unexplored and past work often optimizes
> placement and inference separately. In our work, we develop a deep learning
> approach to optimize both beacon placement and location inference directly
> for localization accuracy. In simulated experiments, our approach
> significantly outperforms strategies that consider beacon placement and
> location inference
>
> separately.
>
>
>
> We then turn our attention to the related problem of task-driven
> optimization of robots and their controllers. Approaches that automate the
> design of robots have a long history and include several techniques such as
> evolutionary algorithms, trajectory optimization, and nonlinear
> programming. Reinforcement learning has proven successful at solving
> complex control problems but, at the start of our work, it was largely
> unexplored for co-optimization. Therefore, we start by proposing a
> data-efficient algorithm based on multi-task reinforcement learning. Our
> approach efficiently optimizes both physical design and control parameters
> directly for task performance by leveraging a design-conditioned controller capable
> of generalizing over the space of physical designs. We then follow this up
> with an extension to allow for the optimization over discrete morphological
> parameters such as the number and configuration of limbs. Finally, we
> conclude by exploring the fabrication and deployment of optimized robots.
> In this work we extend our previous algorithm to allow for the
> co-optimization of soft crawling robots, develop techniques for speeding up finite
> element simulations, and successfully fabricate and transfer the optimized
> robot from simulation to the real world.
>
> *Thesis Committee:*
> *Matthew R. Walter <mwalter at ttic.edu>* *(Thesis Advisor)*
> Ayan Chakrabarti
> Audrey Sedal
> David McAllester
>
>
>
>
>
> Mary C. Marre
> Faculty Administrative Support
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue*
> *Chicago, IL  60637*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20220816/126ff357/attachment-0001.html>