this is a contentImage

A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning(click it !)

Accepted to: AISTATS 2011

Authors: Stéphane Ross, Geoffrey J. Gordon, J. Andrew Bagnell

Leader: Kishan

Venue: Online

Date: 1-2 p.m. May 11st, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Spatial Intention Maps for Multi-Agent Mobile Manipulation(click it !)

Accepted to: ICRA 2021

Authors: Jimmy Wu, Xingyuan Sun, Andy Zeng, Shuran Song, Szymon Rusinkiewicz, Thomas Funkhouser

Leader: Xiaohan

Venue: Online

Date: 1-2 p.m. May 4th, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Continual Learning of Knowledge Graph Embeddings(click it !)

Accepted to: Ra-L 2021

Authors: Angel Daruna, Mehul Gupta, Mohan Sridharan, and Sonia Chernova

Leader: Saeid

Venue: Online

Date: 1-2 p.m. April 20th, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

What Does BERT with Vision Look At?(click it !)

Accepted to: ACL 2020

Authors: Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh and Kai-Wei Chang

Leader: Xiaohan

Venue: Online

Date: 1-2 p.m. April 13th, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Planning with Learned Object Importance in Large Problem Instancesusing Graph Neural Networks(click it !)

Accepted to: AAAI 2021

Authors: Tom Silver, Rohan Chitnis, Aidan Curtis, Joshua Tenenbaum, Tomas Lozano-Perez, Leslie Pack Kaelbling

Leader: Yan

Venue: Online

Date: 1-2 p.m. April 6th, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

A formal methods approach to interpretable reinforcement learning for robotic planning(click it !)

Accepted to: Science Robotics

Authors: Xiao Li, Zachary Serlin, Guang Yang and Calin Belta

Leader: David

Venue: Online

Date: 1-2 p.m. March 30th, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Learning When to Quit: Meta-Reasoning for Motion Planning (click it !)

Accepted to:

Authors: Yoonchang Sung, Leslie Pack Kaelbling, and Tomas Lozano-P'erez

Leader: Yan Ding

Venue: Online

Date: 1-2 p.m. March 23th, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Ethically Compliant Sequential Decision Making (click it !)

Accepted to: AAAI 2021

Authors: Justin Svegliato, Samer B. Nashed, Shlomo Zilberstein

Leader: Saeid

Venue: Online

Date: 1-2 p.m. March 16th, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Hierarchical Planning for Long-Horizon Manipulation with Geometric and Symbolic Scene Graphs (click it !)

Accepted to: ICRA 2021

Authors: Yifeng Zhu, Jonathan Tremblay, Stan Birchfield, Yuke Zhu

Leader: Yan Ding

Venue: Online

Date: 1-2 p.m. March 9th, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Virtual Reality for Robots (click it !)

Accepted to:

Authors: Markku Suomalainen, Alexandra Q. Nilles, and Steven M. LaValle

Leader: Kishan

Venue: Online

Date: 1-2 p.m. March 2nd, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Temporal-Logic-Based Reward Shaping for Continuing Reinforcement Learning Tasks (click it !)

Accepted to: AAAI 2021

Authors: Yuqian Jiang, Suda Bharadwaj, Bo Wu, Rishi Shah, Ufuk Topcu, and Peter Stone

Leader: David

Venue: Online

Date: 1-2 p.m. Feb 23rd, 2021

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Joint Inference of Reward Machines and Policies for Reinforcement Learning (click it !)

Accepted to: ICAPS 2020

Authors: Zhe Xu, Ivan Gavran, Yousef Ahmad, Rupak Majumdar, Daniel Neider, Ufuk Topcu, Bo Wu

Leader: David

Venue: Online

Date: 2-3 p.m. Dec 1st, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

RMM: A Recursive Mental Model for Dialogue Navigation (click it !)

Accepted to: EMNLP(2020)

Authors: Homero Roman Roman, Yonatan Bisk, Jesse Thomason, Asli Celikyilmaz, Jianfeng Gao

Leader: Saeid

Venue: Online

Date: 2-3 p.m. Oct 20th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Human-like Planning for Reaching in Cluttered Environments (click it !)

Accepted to: ICRA 2020

Authors: Mohamed Hasan, Matthew Warburton, Wisdom C. Agboh, Mehmet R. Dogar, Matteo Leonetti, He Wang, Faisal Mushtaq, Mark Mon-Williams, Anthony G. Cohn

Leader: Xiaohan Zhang

Venue: Online

Date: 2-3 p.m. Oct 13th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Making Sense of Vision and Touch: Self-Supervised Learning of multimodal Representations for Contact-Rich Tasks (click it !)

Accepted to: ICRA 2019

Authors: Michelle A. Lee∗, Yuke Zhu∗, Krishnan Srinivasan, Parth Shah, Silvio Savarese, Li Fei-Fei, Animesh Garg, Jeannette Bohg

Leader: Xiaohan Zhang

Venue: Online

Date: 2-3 p.m. Oct 6th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Reasoning About Physical Interactions with Object-Oriented Prediction and Planning (click it !)

Accepted to: ICLR 2019

Authors: Michael Janner, Sergey Levine, William T. Freeman, Joshua B. Tenenbaum, Chelsea Finn, Jiajun Wu

Leader: Shiqi Zhang

Venue: Online

Date: 2-3 p.m. Sep 29th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

How to Build User Simulators to Train RL-based Dialog Systems (click it !)

Accepted to: EMNLP 2019

Authors: Weiyan Shi, Kun Qian, Xuewei Wang, and Zhou Yu

Leader: Xiaohan Zhang

Venue: Online

Date: 2-3 p.m. Sep 22nd, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Semantic Linking Maps for Active Visual Object Search (click it !)

Accepted to: 2020 IEEE International Conference on Robotics and Automation (ICRA)

Authors: Zhen Zeng, Adrian Röfer and Odest Chadwicke Jenkins

Leader: Saeid

Venue: Online

Date: 12-1 p.m. July 9nd, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Simultaneously Learning Transferable Symbols and Language Groundings from Perceptual Data for Instruction Following (click it !)

Accepted to: Robotics: Science and Systems (RSS, 2020)

Authors: Nakul Gopalan, Eric Rosen, George Konidaris, Stefanie Tellex

Leader: Shiqi Zhang

Venue: Online

Date: 12-1 p.m. July 2nd, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

SAIL: Simulation-Informed Active In-the-Wild Learning (click it !)

Accepted to: 2019 14th ACM/IEEE International Conference on Human-Robot Interaction (HRI). IEEE, 2019

Authors: Elaine Schaertl Short; Adam Allevato; Andrea L. Thomaz

Leader: Yan Ding

Venue: Online

Date: 12-1 p.m. June 18th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Improving Grounded Natural Language Understanding through Human-Robot Dialog (click it !)

Authors: Jesse Thomason, Aishwarya Padmakumar, Jivko Sinapov, Nick Walker, Yuqian Jiang, Harel Yedidsion, Justin Hart, Peter Stone, and Raymond J. Mooney

Leader: Xiaohan Zhang

Venue: Online

Date: 12-1 p.m. June 11th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Proximal Policy Optimization Algorithms (click it !)

Authors: John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, Oleg Klimov

Leader: Nicholas A Abate

Venue: Online

Date: 12-1 p.m. June 4th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

RoomShift: Room-scale Dynamic Haptics for VR with Furniture-moving Swarm Robots (click it !)

Authors: Ryo Suzuki, Hooman Hedayati, Clement Zheng, James Bohn, Daniel Szafir, Ellen Yi-Luen Do, Mark D. Gross, Daniel Leithinger

Leader: Kishan D Chandan

Venue: Online

Date: 12-1 p.m. May 21st, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Imagination-Augmented Agents for Deep Reinforcement Learning (click it !)

Authors: Théophane Weber, Sébastien Racanière, David P. Reichert, Lars Buesing, Arthur Guez, Danilo Jimenez Rezende, Adria Puigdomènech Badia, Oriol Vinyals, Nicolas Heess, Yujia Li, Razvan Pascanu, Peter Battaglia, Demis Hassabis, David Silver, Daan Wierstra

Leader: Yan Ding

Venue: Online

Date: 12-1 p.m. May 7th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

That and There: Judging the Intent of Pointing Actions with Robotic Arms (click it !)

Authors: Malihe Alikhani, Baber Khalid, Rahul Shome, Chaitanya MitashKostas Bekris, Matthew Stone

Leader: Yan Ding

Venue: Online

Date: 12-1 p.m. Apr. 23th, 2020

Join Zoom Meeting: https://binghamton.zoom.us/j/3929288199 (click it!)

this is a contentImage

Learning from Interventions using Hierarchical Policies for Safety Learning (click it !)

Presenter:Yan Ding Venue:N09 Date:12-1 p.m. Feb. 20th, 2020

Abstract: Learning from Demonstrations (LfD) via Behavior Cloning (BC) works well on multiple complex tasks. However, a limitation of the typical LfD approach is that it requires expert demonstrations for all scenarios, including those in which the algorithm is already well-trained. The recently proposed Learning from Interventions (LfI) overcomes this limitation by using an expert overseer. The expert overseer only intervenes when it suspects that an unsafe action is about to be taken. Although LfI significantly improves over LfD, the state-of-the-art LfI fails to account for delay caused by the expert's reaction time and only learns short-term behavior. We address these limitations by 1) interpolating the expert's interventions back in time, and 2) by splitting the policy into two hierarchical levels, one that generates sub-goals for the future and another that generates actions to reach those desired sub-goals. This sub-goal prediction forces the algorithm to learn long-term behavior while also being robust to the expert's reaction time. Our experiments show that LfI using sub-goals in a hierarchical policy framework trains faster and achieves better asymptotic performance than typical LfD.

this is a contentImage

Deep Imitation Learning for Autonomous Driving in Generic Urban Scenarios with Enhanced Safety (click it !)

Presenter:Yan Ding, Venue:P03, Date:12-1 p.m. Oct. 23th, 2019

Abstract: The decision and planning system for autonomousdriving in urban environments is hard to design. Most currentmethods manually design the driving policy, which can be ex-pensive to develop and maintain at scale. Instead, with imitation learning we only need to collect data and the computer willlearn and improve the driving policy automatically. However, existing imitation learning methods for autonomous driving arehardly performing well for complex urban scenarios. Moreover,the safety is not guaranteed when we use a deep neural network policy. In this paper, we proposed a framework to learnthe driving policy in urban scenarios efficiently given offline connected driving data, with a safety controller incorporatedto guarantee safety at test time. The experiments show that ourmethod can achieve high performance in realistic simulationsof urban driving scenarios.

this is a contentImage

Robot Programming Through Augmented Trajectories (IROS, 2019) (click it !)

Presenter:Kishan Chandan, Venue:P03 Date:12-1 p.m., Oct. 16th, 2019

Abstract: This paper presents a future-focused approach for robot programming based on augmented trajectories. Using a mixed reality head-mounted display (Microsoft Hololens) and a 7-DOF robot arm, we designed an augmented reality (AR) robotic interface with four interactive functions to ease the robot programming task: 1) Trajectory specification. 2) Virtual previews of robot motion. 3) Visualization of robot parameters. 4) Online reprogramming during simulation and execution. We validate our AR-robot teaching interface by comparing it with a kinesthetic teaching interface in two different scenarios as part of a pilot study: creation of contact surface path and free space path. Furthermore, we present an industrial case study that illustrates our AR manufacturing paradigm by interacting with a 7-DOF robot arm to reduce wrinkles during the pleating step of the carbon-fiber-reinforcement-polymer vacuum bagging process in a simulated scenario.

this is a contentImage

Learning to Teach in Cooperative Multiagent Reinforcement Learning (AAAI, 2019) (click it !)

Presenters:Shiqi Zhang and Saeid Amiri, Venue:G11 Date:12-1 p.m., Apr. 22th, 2019

Abstract: Collective human knowledge has clearly benefited from thefact that innovations by individuals are taught to others throughcommunication. Similar to human social groups, agents indistributed learning systems would likely benefit from com-munication to share knowledge and teach skills. The problemof teaching to improve agent learning has been investigatedby prior works, but these approaches make assumptions thatprevent application of teaching to general multiagent prob-lems, or require domain expertise for problems they can applyto. This learning to teach problem has inherent complexitiesrelated to measuring long-term impacts of teaching that com-pound the standard multiagent coordination challenges. Incontrast to existing works, this paper presents the first gen-eral framework and algorithm for intelligent agents to learn toteach in a multiagent environment. Our algorithm, Learningto Coordinate and Teach Reinforcement (LeCTR), addressespeer-to-peer teaching in cooperative multiagent reinforcementlearning. Each agent in our approach learns both when andwhat to advise, then uses the received advice to improve locallearning. Importantly, these roles are not fixed; these agentslearn to assume the role of student and/or teacher at the ap-propriate moments, requesting and providing advice in orderto improve teamwide performance and learning. Empiricalcomparisons against state-of-the-art teaching methods showthat our teaching agents not only learn significantly faster, butalso learn to coordinate in tasks where existing methods fail.

this is a contentImage

Using Natural Language for Reward Shaping in Reinforcement Learning (click it !)

Presenter:Saeid Amiri, Venue:G11 Date:12-1 p.m., Apr. 15th, 2019

Abstract: Recent reinforcement learning (RL) approacheshave shown strong performance in complex do-mains such as Atari games, but are often highlysample inefficient.A common approach to re-duce interaction time with the environment is touse reward shaping, which involves carefully de-signing reward functions that provide the agent in-termediate rewards for progress towards the goal.However, designing appropriate shaping rewards isknown to be difficult as well as time-consuming. Inthis work, we address this problem by using naturallanguage instructions to perform reward shaping.We propose the LanguagE-Action Reward Network(LEARN), a framework that maps free-form nat-ural language instructions to intermediate rewardsbased on actions taken by the agent. These inter-mediate language-based rewards can seamlessly beintegrated into any standard reinforcement learn-ing algorithm. We experiment with Montezuma’sRevenge from the Atari Learning Environment, apopular benchmark in RL. Our experiments on adiverse set of 15 tasks demonstrate that, for thesame number of interactions with the environment,language-based rewards lead to successful comple-tion of the task 60% more often on average, com-pared to learning without language.

this is a contentImage

Agile Autonomous Driving using End-to-End Deep Imitation Learning (RSS, 2018) (click it!)

Presenter:Roger Correia, Venue:P03 Date:12-1 p.m., Apr. 1st, 2019

Abstract: We present an end-to-end imitation learning system for agile, off-road autonomous driving using only low-cost sensors. By imitating a model predictive controller equipped with advanced sensors, we train a deep neural network control policy to map raw, high-dimensional observations to continuous steering and throttle commands. Compared with recent approaches to similar tasks, our method requires neither state estimation nor on-the-fly planning to navigate the vehicle. Our approach relies on, and experimentally validates, recent imitation learning theory. Empirically, we show that policies trained with online imitation learning overcome well-known challenges related to covariate shift and generalize better than policies trained with batch imitation learning. Built on these insights, our autonomous driving system demonstrates successful high-speed off-road driving, matching the state-of-the-art performance.

this is a contentImage

Adversarial Actor-Critic Method for Task and Motion Planning Problems Using Planning Experience (AAAI, 2019) (click it !)

Presenter:Roger Correia Venue:G11 Date:12-1 p.m. Mar. 25th, 2019

Abstract: We propose an actor-critic algorithm that uses past plan-ning experience to improve the efficiency of solving robottask-and-motion planning (TAMP) problems. TAMP plan-ners search for goal-achieving sequences of high-level op-erator instances specified by both discrete and continuousparameters. Our algorithm learns a policy for selecting thecontinuous parameters during search, using a small trainingset generated from the search trees of previously solved in-stances. We also introduce a novel fixed-length vector rep-resentation for world states with varying numbers of objectswith different shapes, based on a set of key robot configura-tions. We demonstrate experimentally that our method learnsmore efficiently from less data than standard reinforcement-learning approaches and that using a learned policy to guidea planner results in the improvement of planning efficiency.

this is a contentImage

A Network-based End-to-End Trainable Task-oriented Dialogue System (Association for Computational Linguistic, 2017) (click it !)

Presenter:Shiqi Zhang Venue:G11 Date:12-1 p.m. Mar. 11th, 2019

Abstract: Teaching machines to accomplish tasksby conversing naturally with humans ischallenging. Currently, developing task-oriented dialogue systems requires creatingmultiple components and typically this in-volves either a large amount of handcraft-ing, or acquiring costly labelled datasetsto solve a statistical learning problem foreach component. In this work we intro-duce a neural network-based text-in, text-out end-to-end trainable goal-oriented di-alogue system along with a new way ofcollecting dialogue data based on a novelpipe-lined Wizard-of-Oz framework. Thisapproach allows us to develop dialogue sys-tems easily and without making too manyassumptions about the task at hand. Theresults show that the model can conversewith human subjects naturally whilst help-ing them to accomplish tasks in a restaurantsearch domain.

this is a contentImage

Communicating Robot Motion Intent with Augmented Reality (HRI, 2018) (click it !)

Presenter:Kishan Chandan Venue:G11 Date:12-1 p.m. Mar. 4th, 2019

Abstract: Humans coordinate teamwork by conveying intent through social cues, such as gestures and gaze behaviors. However, these methods may not be possible for appearance-constrained robots that lack anthropomorphic or zoomorphic features, such as aerial robots. We explore a new design space for communicating robot motion intent by investigating how augmented reality (AR) might mediate human-robot interactions. We develop a series of explicit and implicit designs for visually signaling robot motion intent using AR, which we evaluate in a user study. We found that several of our AR designs significantly improved objective task efficiency over a baseline in which users only received physically-embodied orientation cues. In addition, our designs offer several trade-offs in terms of intent clarity and user perceptions of the robot as a teammate.

this is a contentImage

Learning Pipelines with Limited Data and Domain Knowledge (NeurIPS, 2018) (click it !)

Presenter:Saeid Amiri, Venue:G11 Date:12-1 p.m. Feb. 11st, 2019

Abstract: As machine learning becomes more widely used in practice, we need new methods to build complex intelligent systems that integrate learning with existing software,and with domain knowledge encoded as rules. As a case study, we present sucha system that learns to parse Newtonian physics problems in textbooks. This system, Nuts&Bolts, learns a pipeline process that incorporates existing code,pre-learned machine learning models, and human engineered rules. It jointly trainsthe entire pipeline to prevent propagation of errors, using a combination of labelledand unlabelled data. Our approach achieves a good performance on the parsingtask, outperforming the simple pipeline and its variants. Finally, we also show how Nuts&Boltscan be used to achieve improvements on a relation extraction taskand on the end task of answering Newtonian physics problems.

this is a contentImage

Behavioral Cloning from Observation (IJCAI, 2018) (click it !)

Presenter:Saeid Amiri, Venue:G11 Date:12-1 p.m. Feb. 11st, 2019

Abstract: Humans often learn how to perform tasks via imitation: they observe others perform a task, and then very quickly infer the appropriate actions to take based on their observations. While extending this paradigm to autonomous agents is a well-studied problem in general, there are two particular aspects that have largely been overlooked: (1) that the learning is done from observation only (i.e., without explicit action information), and (2) that the learning is typically done very quickly. In this work, we propose a two-phase, autonomous imitation learning technique called behavioral cloning from observation (BCO), that aims to provide improved performance with respect to both of these aspects. First, we allow the agent to acquire experience in a self-supervised fashion. This experience is used to develop a model which is then utilized to learn a particular task by observing an expert perform that task without the knowledge of the specific actions taken. We experimentally compare BCO to imitation learning methods, including the state-of-the-art, generative adversarial imitation learning (GAIL) technique, and we show comparable task performance in several different simulation domains while exhibiting increased learning speed after expert trajectories become available.

About Reading Group

  • Reading group hopes to broaden the research scope of students, especially in human-robot system, by discussing good new papers. This meeting will be held every week by AIR Group at SUNY Binghamton. If you like discussing interesting research ideas, please join HRS Reading Group without any hesitation!

Branches

How to join?

| totalview: | visitors: