Liam Paull


Some Current Projects

prius


CS
AIL/Toyota Autonomous Driving Project: An initiative launched in Jan. 2016 to build a car - driven by a human - but incapable of crashing.

[planning paper] [systems paper]


Duckietown: What is Duckietown? It is: a fictional startup, a class taught at MIT, a research endeavor, a rebranding of robotics, an outreach effort, and an opportunity for all of us to learn something.

[research paper] [education paper] [code] [slides] 


Unified Approach to Resource-Constrained Graph SLAM
: In this work we generalize the previous work on communication-constrained inference with graphical methods to incorporate general resource constraints such as limited memory or computation.

[paper] [code] [slides] [poster]


Some Old Projects



SLAM with Objects Using a Nonparametric Pose Graph
(with Beipeng Mu): We leverage Bayesian non-parametrics to cluster object detections into distinct groups so that they can be used as features in a pose graph.

[paper] [video]


Information-based Active SLAM via Topological Feature Graphs
(with Beipeng Mu): We move away from the occupancy grid map representation that is standard in active SLAM problem in a unified way over a new data structure: the topological feature graph.

[paper] [video]


"Focused" Inference for Resource-Constrained Navigation
(with Beipeng Mu): In this work we propose a two-stage process for selecting landmarks and measurements to support a navigation task in the presence of resource constraints

[TRO paper] [RSS paper] [slides] [poster] [video] [code]
Cooperative Coverage


Cooperative Communication-Constrained Area Coverage
: In previous work we have linked area coverage and localization. Now we exploit the benefits of underwater cooperative localization to improve our ability to perform coverage. Experiments are performed using the HoverGroup's autonomous kayaks.

[paper] [workshop paper] [slides] [video] [code]
Cooperative AUV SLAM


Cooperative Communication-Constrained AUV SLAM: The acoustic communications channel is simply too restrictive to pass all feature information between submerged autonomous underwater vehicles. We develop a principled framework for how and what to transmit so that vehicles can cooperatively map and localize concurrently.

[paper] [slides] [video] [code]


Text Spotting and SLAM
(with Hsueh-Cheng Wang): We use SLAM to improve environmental text identification, and use text to as features in the SLAM system to improve localization.

[paper] [video] [slides]