[Colloquium] REMINDER: 1/20 Talks at TTIC: David Rosen, MIT

[Mary Marre]

*When:*      Wednesday, January 20th at* 11:10 am CT*



*Where:*     Zoom Virtual Talk (*register in advance here
<https://uchicagogroup.zoom.us/webinar/register/WN_pe1CHU7SSnWzn7-xTvdbHg>*)



*Who: *       David Rosen, MIT

*Title:*  Provably Sound Perception for Reliable Autonomy
*Abstract:*  Machine perception -- the ability to construct accurate models
of the world from raw sensor data -- is an essential capability for mobile
robots, supporting such fundamental functions as planning, navigation, and
control.  However, the development of algorithms for robotic perception
that are both *practical* and *reliable* presents a formidable challenge:
many fundamental perception tasks require the solution of computationally
hard estimation problems, yet practical methods are constrained to run in
real-time on resource-limited mobile platforms. Moreover, reliable
perception algorithms must also be robust to the myriad challenges
encountered in real-world operation, including sensor noise, uncertain or
misspecified perceptual models, and potentially corrupted data (from e.g.
sensor faults).

In this talk, I present my work on the design of practical provably sound
machine perception algorithms, focusing on the motivating application of
spatial perception.  First, I address the fundamental problem of pose-graph
optimization (PGO); this is a high-dimensional estimation problem over a
nonconvex state space, and so is computationally challenging to solve in
general.  Nevertheless, we present a convex relaxation whose minimizer
provides an *exact, globally optimal* PGO solution in a noise regime that
includes most practical applications in robotics and computer vision.  We
leverage this relaxation to develop SE-Sync, the first practical SLAM
algorithm provably capable of recovering *certifiably correct* solutions.

Second, I briefly describe our recent work on learning to estimate
rotations.  We show that topological obstructions prevent deep neural
networks (DNNs) employing common rotation parameterizations (e.g.
quaternions) from learning to estimate widely-dispersed rotation targets.
We describe a novel parameterization of 3D rotations that overcomes this
obstruction, and that supports an explicit notion of uncertainty in our
networks’ predictions.  We show that DNNs employing our representation are
consistently more accurate when applied to object pose estimation tasks,
and that their predicted uncertainties enable the reliable identification
of out-of-distribution test examples (including corrupted inputs).

Finally, I will conclude with a discussion of future directions that aim to
unify provably sound estimation and learning methods, thereby enabling the
creation of perception systems with both the *robustness* and
*adaptability* necessary to support reliable long-term robotic autonomy in
the real world.

*Bio:  *David M. Rosen is a postdoctoral associate in the Laboratory for
Information and Decision Systems at the Massachusetts Institute of
Technology.  Prior to joining LIDS, he was a Research Scientist at Oculus
Research (now Facebook Reality Labs) in Seattle.  He received his ScD in
Computer Science from the Massachusetts Institute of Technology in 2016.
His research addresses the design of practical provably sound methods for
machine perception, using a combination of tools from optimization,
geometry, algebra, and probabilistic inference. His work has been
recognized with a Best Paper Award at the 2016 International Workshop on
the Algorithmic Foundations of Robotics, an RSS Pioneer Award at Robotics:
Science and Systems 2019, and a Best Student Paper Award at Robotics:
Science and Systems 2020.

*Host:* Matthew Walter <mwalter at ttic.edu>



Mary C. Marre
Faculty Administrative Support
*Toyota Technological Institute*
*6045 S. Kenwood Avenue*
*Room 517*
*Chicago, IL  60637*
*p:(773) 834-1757*
*f: (773) 357-6970*
*mmarre at ttic.edu <mmarre at ttic.edu>*


On Tue, Jan 19, 2021 at 12:36 PM Mary Marre <mmarre at ttic.edu> wrote:

>
> *When:*      Wednesday, January 20th at* 11:10 am CT*
>
>
>
> *Where:*     Zoom Virtual Talk (*register in advance here
> <https://uchicagogroup.zoom.us/webinar/register/WN_pe1CHU7SSnWzn7-xTvdbHg>*
> )
>
>
>
> *Who: *       David Rosen, MIT
>
> *Title:*  Provably Sound Perception for Reliable Autonomy
> *Abstract:*  Machine perception -- the ability to construct accurate
> models of the world from raw sensor data -- is an essential capability for
> mobile robots, supporting such fundamental functions as planning,
> navigation, and control.  However, the development of algorithms for
> robotic perception that are both *practical* and *reliable* presents a
> formidable challenge: many fundamental perception tasks require the
> solution of computationally hard estimation problems, yet practical methods
> are constrained to run in real-time on resource-limited mobile platforms.
> Moreover, reliable perception algorithms must also be robust to the myriad
> challenges encountered in real-world operation, including sensor noise,
> uncertain or misspecified perceptual models, and potentially corrupted data
> (from e.g. sensor faults).
>
> In this talk, I present my work on the design of practical provably sound
> machine perception algorithms, focusing on the motivating application of
> spatial perception.  First, I address the fundamental problem of pose-graph
> optimization (PGO); this is a high-dimensional estimation problem over a
> nonconvex state space, and so is computationally challenging to solve in
> general.  Nevertheless, we present a convex relaxation whose minimizer
> provides an *exact, globally optimal* PGO solution in a noise regime that
> includes most practical applications in robotics and computer vision.  We
> leverage this relaxation to develop SE-Sync, the first practical SLAM
> algorithm provably capable of recovering *certifiably correct* solutions.
>
> Second, I briefly describe our recent work on learning to estimate
> rotations.  We show that topological obstructions prevent deep neural
> networks (DNNs) employing common rotation parameterizations (e.g.
> quaternions) from learning to estimate widely-dispersed rotation targets.
> We describe a novel parameterization of 3D rotations that overcomes this
> obstruction, and that supports an explicit notion of uncertainty in our
> networks’ predictions.  We show that DNNs employing our representation are
> consistently more accurate when applied to object pose estimation tasks,
> and that their predicted uncertainties enable the reliable identification
> of out-of-distribution test examples (including corrupted inputs).
>
> Finally, I will conclude with a discussion of future directions that aim
> to unify provably sound estimation and learning methods, thereby enabling
> the creation of perception systems with both the *robustness* and
> *adaptability* necessary to support reliable long-term robotic autonomy in
> the real world.
>
> *Bio:  *David M. Rosen is a postdoctoral associate in the Laboratory for
> Information and Decision Systems at the Massachusetts Institute of
> Technology.  Prior to joining LIDS, he was a Research Scientist at Oculus
> Research (now Facebook Reality Labs) in Seattle.  He received his ScD in
> Computer Science from the Massachusetts Institute of Technology in 2016.
> His research addresses the design of practical provably sound methods for
> machine perception, using a combination of tools from optimization,
> geometry, algebra, and probabilistic inference. His work has been
> recognized with a Best Paper Award at the 2016 International Workshop on
> the Algorithmic Foundations of Robotics, an RSS Pioneer Award at Robotics:
> Science and Systems 2019, and a Best Student Paper Award at Robotics:
> Science and Systems 2020.
>
> *Host:* Matthew Walter <mwalter at ttic.edu>
>
>
>
>
> Mary C. Marre
> Faculty Administrative Support
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue*
> *Room 517*
> *Chicago, IL  60637*
> *p:(773) 834-1757*
> *f: (773) 357-6970*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
>
> On Wed, Jan 13, 2021 at 8:02 PM Mary Marre <mmarre at ttic.edu> wrote:
>
>>
>> *When:*      Wednesday, January 20th at* 11:10 am CT*
>>
>>
>>
>> *Where:*     Zoom Virtual Talk (*register in advance here
>> <https://uchicagogroup.zoom.us/webinar/register/WN_pe1CHU7SSnWzn7-xTvdbHg>*
>> )
>>
>>
>>
>> *Who: *       David Rosen, MIT
>>
>> *Title:*  Provably Sound Perception for Reliable Autonomy
>> *Abstract:*  Machine perception -- the ability to construct accurate
>> models of the world from raw sensor data -- is an essential capability for
>> mobile robots, supporting such fundamental functions as planning,
>> navigation, and control.  However, the development of algorithms for
>> robotic perception that are both *practical* and *reliable* presents a
>> formidable challenge: many fundamental perception tasks require the
>> solution of computationally hard estimation problems, yet practical methods
>> are constrained to run in real-time on resource-limited mobile platforms.
>> Moreover, reliable perception algorithms must also be robust to the myriad
>> challenges encountered in real-world operation, including sensor noise,
>> uncertain or misspecified perceptual models, and potentially corrupted data
>> (from e.g. sensor faults).
>>
>> In this talk, I present my work on the design of practical provably sound
>> machine perception algorithms, focusing on the motivating application of
>> spatial perception.  First, I address the fundamental problem of pose-graph
>> optimization (PGO); this is a high-dimensional estimation problem over a
>> nonconvex state space, and so is computationally challenging to solve in
>> general.  Nevertheless, we present a convex relaxation whose minimizer
>> provides an *exact, globally optimal* PGO solution in a noise regime that
>> includes most practical applications in robotics and computer vision.  We
>> leverage this relaxation to develop SE-Sync, the first practical SLAM
>> algorithm provably capable of recovering *certifiably correct* solutions.
>>
>> Second, I briefly describe our recent work on learning to estimate
>> rotations.  We show that topological obstructions prevent deep neural
>> networks (DNNs) employing common rotation parameterizations (e.g.
>> quaternions) from learning to estimate widely-dispersed rotation targets.
>> We describe a novel parameterization of 3D rotations that overcomes this
>> obstruction, and that supports an explicit notion of uncertainty in our
>> networks’ predictions.  We show that DNNs employing our representation are
>> consistently more accurate when applied to object pose estimation tasks,
>> and that their predicted uncertainties enable the reliable identification
>> of out-of-distribution test examples (including corrupted inputs).
>>
>> Finally, I will conclude with a discussion of future directions that aim
>> to unify provably sound estimation and learning methods, thereby enabling
>> the creation of perception systems with both the *robustness* and
>> *adaptability* necessary to support reliable long-term robotic autonomy in
>> the real world.
>>
>> *Bio:  *David M. Rosen is a postdoctoral associate in the Laboratory for
>> Information and Decision Systems at the Massachusetts Institute of
>> Technology.  Prior to joining LIDS, he was a Research Scientist at Oculus
>> Research (now Facebook Reality Labs) in Seattle.  He received his ScD in
>> Computer Science from the Massachusetts Institute of Technology in 2016.
>> His research addresses the design of practical provably sound methods for
>> machine perception, using a combination of tools from optimization,
>> geometry, algebra, and probabilistic inference. His work has been
>> recognized with a Best Paper Award at the 2016 International Workshop on
>> the Algorithmic Foundations of Robotics, an RSS Pioneer Award at Robotics:
>> Science and Systems 2019, and a Best Student Paper Award at Robotics:
>> Science and Systems 2020.
>>
>> *Host:* Matthew Walter <mwalter at ttic.edu>
>>
>>
>> Mary C. Marre
>> Faculty Administrative Support
>> *Toyota Technological Institute*
>> *6045 S. Kenwood Avenue*
>> *Room 517*
>> *Chicago, IL  60637*
>> *p:(773) 834-1757*
>> *f: (773) 357-6970*
>> *mmarre at ttic.edu <mmarre at ttic.edu>*
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20210120/2c7cd4de/attachment-0001.html>