thefluentroboticslab

HOUND: An open-source, low-cost research platform for high-speed off-road underactuated nonholonomic driving

We present HOUND, a small-scale, inexpensive, off-road autonomy platform that handles challenging off-road terrains at high speeds through the integration of a rollover prevention system. HOUND is integrated with a complete stack for perception and control.

2024

Cook2LTL: Translating cooking recipes to LTL formulae using large language models

We present a system that translates natural language instructions to sequences of robot-executable robot commands in the form of linear temporal logic formulae. We demonstrate the system in the domain of cooking recipes through simulations in AI2-THOR.

Abstracting road traffic via topological braids: Applications to traffic flow analysis and distributed control

We abstract road traffic using topological braids. We show that braids may uncover insights about the interactivity of vehicles in complex scenes, and guide decision making of non-communicating agents.

Electrostatic brakes enable individual joint control of underactuated, highly articulated robots

We developed a lightweight, power-efficient mechanism based on electrostatic-braking technology. Using only 1 motor, our mechanism enabled a 10-DoF manipulator to perform a complex multi-object manipulation task.

2023

From crowd motion prediction to robot navigation in crowds

We investigate the transfer of state-of-the-art human motion prediction models to the crowd navigation performance of a mobile robot.

PuSHR: A multirobot system for nonprehensile rearrangement

Using MuSHR racecars, we built a multirobot system that combines task assignment, multiagent motion planning, and decentralized control to automatically rearrange a cluttered workspace.

Winding through: Crowd navigation via topological invariance

We formalize passing between two agents as a rotation, using a notion of topological invariance, and build a MPC that expedites passing progress in crowded spaces. We demonstrate our approach on a self-balancing robot navigating among 3 users in a lab experiment.

Core challenges of social robot navigation: A survey

We organize challenges in social robot navigation into a set of categories related to broader open problems in robot planning, behavior design, and evaluation methodologies. We review past work, diagnose limitations of adopted practices, and offer constructive feedback and directions for future research.

2022

Analyzing multiagent interactions in traffic scenes via topological braids

We show that topological braids can be used as a tool to characterize the interactivity of vehicles, which can be useful for evaluating the richness of existing datasets, but also for synthetically designing balanced training datasets and benchmarks.

Not all who wander are lost: A localization-free system for in-the-wild mobile robot deployments

We developed a new gripper with optical proximity sensing on fingertips. We demonstrate that proximity sensing enables significantly more accurate object pose estimattion during manipulation compared to tactile sensing.

Optical proximity sensing for pose estimation during in-hand manipulation

We developed a new gripper with optical proximity sensing on fingertips. We demonstrate that proximity sensing enables significantly more accurate object pose estimattion during manipulation compared to tactile sensing.

Social Momentum: Design and evaluation of a framework for socially competent robot navigation

We summarize our findings from two large-scale user studies evaluating Social Momentum, a reactive controller for robot crowd navigation. We provide experimental insights and lessons learned for algorithmic and experimental design for social robot navigation.

2021

Group-based motion prediction for navigation in crowded environments

Inspired by Gestalt theories from psychology, we build a group-based prediction model and integrate it into a MPC, demonstrating higher safety and fewer group intrusions compared to baselines using individual motion prediction across extensive simulations and real-world pedestrian datasets.

Influencing behavioral attributions to robot motion during task execution

We propose a framework that probabilistically maps robot motion to user attributions and use it to generate robot motion that elicits desired attributions while completing a robot vacuum cleaning task.

Modeling human helpfulness with individual and contextual factors for robot planning

We develop a model of human helpfulness towards a robot that accounts for individual and contextual factors and integrate it in a BAMDP policy that empowers a robot to get more help in a localization task situated in a virtual office environment.

Hamiltonian coordination primitives for decentralized multiagent navigation

We present a framework that generates multiagent trajectories satisfying desired topological specifications and use it to build a decentralized control policy for navigation in multiagent domains.

2020

Multimodal trajectory prediction via topological invariance for navigation at uncontrolled intersections

We abstract traffic at intersections into a set of modes using a notion of topological invariance, and construct a Graph Neural Network based model (MTP) that conditions multiagent trajectory prediction on the likelihood of modes given observation of past behavior. MTP empowers a MPC to safely traverse uncontrolled intersections across a variety of challenging intersection scenarios on the CARLA simulator.

Telemanipulation with chopsticks: Analyzing human factors in user demonstrations

We developed a new system for chopsticks-based telemanipulation and conducted a user study (N=25) to gain insights into human manipulation strategies. We show that our system enables novice users to complete a series of challenging grasping tasks.

2019

Multi-agent path topology in support of socially competent navigation planning

We introduce the idea of using topological braids as representatives of coordination strategies braids in multiagent navigation domains. Simulated experiments in a gridworld show that agents reasoning about braids may coordinate efficiently without relying on explicit communication.

MuSHR: A low-cost, open-source robotic racecar for education and research

We present MuSHR, the Multi-agent System for non-Holonomic Racing. MuSHR is a low-cost, open-source robotic racecar for education and research, aspiring to democratize robotics via documentation, tutorials, demos, and courses.

2017

Socially competent navigation planning by deep learning of multi-agent path topologies

We use a LSTM-based architecture to learn a mapping from trajectories to a probability distribution over multiagent navigation strategies, encoded as topological braids. We show that this model enables non-communicating agents to coordinate efficienty across a seris of challenging simulated scenarios.

Implicit communication in a joint action

We formalize a general framework that enables robots to encode communicative signals on their actions to seamlessly convey information to their human partners on top of accomplishing the functional objectives of a task. We show how the framework maps onto various modalities, including natural language and motion.

2015

Open-source, anthropomorphic, underactuated robot hands with a selectively lockable differential mechanism: Towards affordable prostheses

We present a new, open-source mechanical hand, designed according to human anthropometry studies to support the development of low-cost prosthetics. The hand can be assembled using off-the-shelf components and parts fabricated via standard prototyping techniques.

Quantifying anthropomorphism of robot arms

We introduce an index for assessing the anthropomorphism of a robot manipulator by comparing its workspace to the workspace of a human arm. A case study on commercial manipulators demonstrates the value of the approach.

2014

Task-specific grasp selection for underactuated hands

We propose an optimization-based approach for deriving task-specific, force-closure grasps for underactuated robot hands. We demonstrate the efficiency of the approach through simulations with the DLR/HIT II five-fingered robot hand.

Open-Source, Affordable, Modular, Light-Weight, Underactuated Robot Hands

We present a new underactuated robot hand that can be built using off-the-shelf components, and easily fabricated parts. Real-world experiments demonstrate the promise of the design for grasping a wide range of everyday-life objects.

2013

Sequential improvement of grasp based on sensitivity analysis

We present an algorithm that takes as input a locally optimal robot grasp and outputs a grasp of higher quality by sequentially perturbing the hand posture and force exertion via sensitivity analysis. Simulations on the DLR/HIT Hand II demonstrate the value of the approach.