Thomas Kollar

EDBT 2026 Demo / reviewers in the wild / expert

dblp:10/6653 · DBLP profile ↗

← Back

39ranked-venue papers

6as first author

19since 2021 · last 2025

0000-0003-2598-8118ORCID · corroborated

Domains — the database's venue-derived domains; a paper can count in several

Artificial intelligence and machine learning · 38 · 6 first-author · 19 since 2021Systems, architecture and hardware · 21 · 3 first-author · 9 since 2021Graphics, computer vision, multimedia, augmented reality and games · 7 · 1 first-author · 3 since 2021Human-computer interaction and ubiquitous computing · 3 · 2 first-author

Year	Publication	Venue	Position
2025	Language models scale reliably with over-training and on downstream tasks abstract Scaling laws are useful guides for derisking expensive training runs, as they predict performance of large models using cheaper, small-scale experiments. However, there remain gaps between current scaling studies and how language models are ultimately trained and evaluated. For instance, scaling is usually studied in the compute-optimal training regime (i.e., "Chinchilla optimal" regime). In contrast, models are often over-trained to reduce inference costs. Moreover, scaling laws mostly predict loss on next-token prediction, but models are usually compared on downstream task performance. To address both shortcomings, we create a testbed of 104 models with 0.011B to 6.9B parameters trained with various numbers of tokens on three data distributions. First, we fit scaling laws that extrapolate in both the amount of over-training and the number of model parameters. This enables us to predict the validation loss of a 1.4B parameter, 900B token run (i.e., 32$\times$ over-trained) and a 6.9B parameter, 138B token run (i.e., a compute-optimal run)––each from experiments that take 300$\times$ less compute. Second, we relate the perplexity of a language model to its downstream task performance by proposing a power law. We use this law to predict top-1 error averaged over downstream tasks for the two aforementioned models, using experiments that take 20$\times$ less compute. Samir Yitzhak Gadre, Georgios Smyrnis, Vaishaal Shankar, Suchin Gururangan, Mitchell Wortsman, Rulin Shao, Jean Mercat, Alex Fang, Jeffrey Li, Sedrick Keh, Marianna Nezhurina, Igor Vasiljevic, Luca Soldaini, Jenia Jitsev, Alexandros G. Dimakis, Gabriel Ilharco, Pang Wei Koh, Shuran Song, Thomas Kollar	ICLR	20
2025	Should VLMs be Pre-trained with Image Data? abstract Pre-trained LLMs that are further trained with image data perform well on vision-language tasks. While adding images during a second training phase effectively unlocks this capability, it is unclear how much of a gain or loss this two-step pipeline gives over VLMs which integrate images earlier into the training process. To investigate this, we train models spanning various datasets, scales, image-text ratios, and amount of pre-training done before introducing vision tokens. We then fine-tune these models and evaluate their downstream performance on a suite of vision-language and text-only tasks. We find that pre-training with a mixture of image and text data allows models to perform better on vision-language tasks while maintaining strong performance on text-only evaluations. On an average of 6 diverse tasks, we find that for a 1B model, introducing visual tokens 80\% of the way through pre-training results in a 2\% average improvement over introducing visual tokens to a fully pre-trained model. Sedrick Keh, Jean Mercat, Samir Yitzhak Gadre, Kushal Arora, Igor Vasiljevic, Benjamin Burchfiel, Shuran Song, Russ Tedrake, Thomas Kollar, Ludwig Schmidt, Achal Dave	ICLR	9
2025	Understanding Complexity in VideoQA via Visual Program Generation abstract We propose a data-driven approach to analyzing query complexity in Video Question Answering (VideoQA). Previous efforts in benchmark design have relied on human expertise to design challenging questions, yet we experimentally show that humans struggle to predict which questions are difficult for machine learning models. Our automatic approach leverages recent advances in code generation for visual question answering, using the complexity of generated code as a proxy for question difficulty. We demonstrate that this measure correlates significantly better with model performance than human estimates. To operationalize this insight, we propose an algorithm for estimating question complexity from code. It identifies fine-grained primitives that correlate with the hardest questions for any given set of models, making it easy to scale to new approaches in the future. Finally, to further illustrate the utility of our method, we extend it to automatically generate complex questions, constructing a new benchmark that is 1.9 times harder than the popular NExT-QA. Cristóbal Eyzaguirre, Igor Vasiljevic, Achal Dave, Jiajun Wu 0001, Rares Ambrus, Thomas Kollar, Juan Carlos Niebles, Pavel Tokmakov	ICML	6
2025	GHIL-Glue: Hierarchical Control with Filtered Subgoal Images abstract Image and video generative models that are pretrained on Internet-scale data can greatly increase the generalization capacity of robot learning systems. These models can function as high-level planners, generating intermediate sub-goals for low-level goal-conditioned policies to reach. However, the performance of these systems can be greatly bottlenecked by the interface between generative models and low-level controllers. For example, generative models may predict photo-realistic yet physically infeasible frames that confuse low-level policies. Low-level policies may also be sensitive to subtle visual artifacts in generated goal images. This paper addresses these two facets of generalization, providing an interface to effectively “glue together” language-conditioned image or video prediction models with low-level goal-conditioned policies. Our method, Generative Hierarchical Imitation Learning-Glue (GHIL-Glue), filters out subgoals that do not lead to task progress and improves the robustness of goal-conditioned policies to generated subgoals with harmful visual artifacts. We find in extensive experiments in both simulated and real environments that GHIL-Glue achieves a 25% improvement across several hierarchical models that leverage generative subgoals, achieving a new state-of-the-art on the CALVIN simulation benchmark for policies using observations from a single RGB camera. GHIL-Glue also outperforms other generalist robot policies across 3/4 language-conditioned manipulation tasks testing zero-shot generalization in physical experiments. Code, model checkpoints, videos, and supplementary materials can be found at https://ghil-glue.github.io. Kyle Hatch, Ashwin Balakrishna, Oier Mees, Suraj Nair 0003, Seohong Park, Blake Wulfe, Masha Itkina, Benjamin Eysenbach, Sergey Levine, Thomas Kollar, Benjamin Burchfiel	ICRA	10
2024	Prismatic VLMs: Investigating the Design Space of Visually-Conditioned Language Models abstract Visually-conditioned language models (VLMs) have seen growing adoption in applications such as visual dialogue, scene understanding, and robotic task planning; adoption that has fueled a wealth of new models such as LLaVa, InstructBLIP, and PaLI-3. Despite the volume of new releases, key design decisions around image preprocessing, architecture, and optimization are under-explored, making it challenging to understand what factors account for model performance – a challenge further complicated by the lack of objective, consistent evaluations. To address these gaps, we first compile a suite of standardized evaluations spanning visual question answering, object localization, and challenge sets that probe properties such as hallucination; evaluations that provide fine-grained insight VLM capabilities. Second, we rigorously investigate VLMs along key design axes, including pretrained visual representations and training from base vs. instruct-tuned language models, amongst others. We couple our analysis with three resource contributions: (1) a unified framework for evaluating VLMs, (2) optimized, flexible training code, and (3) checkpoints for all models, including a family of VLMs at the 7-13B scale that strictly outperform InstructBLIP and LLaVa v1.5, the state-of-the-art in open VLMs. Siddharth Karamcheti, Suraj Nair 0003, Ashwin Balakrishna, Percy Liang, Thomas Kollar, Dorsa Sadigh	ICML	5
2024	Open X-Embodiment: Robotic Learning Datasets and RT-X Models : Open X-Embodiment Collaboration abstract Large, high-capacity models trained on diverse datasets have shown remarkable successes on efficiently tackling downstream applications. In domains from NLP to Computer Vision, this has led to a consolidation of pretrained models, with general pretrained backbones serving as a starting point for many applications. Can such a consolidation happen in robotics? Conventionally, robotic learning methods train a separate model for every application, every robot, and even every environment. Can we instead train "generalist" X-robot policy that can be adapted efficiently to new robots, tasks, and environments? In this paper, we provide datasets in standardized data formats and models to make it possible to explore this possibility in the context of robotic manipulation, alongside experimental results that provide an example of effective X-robot policies. We assemble a dataset from 22 different robots collected through a collaboration between 21 institutions, demonstrating 527 skills (160266 tasks). We show that a high-capacity model trained on this data, which we call RT-X, exhibits positive transfer and improves the capabilities of multiple robots by leveraging experience from other platforms. The project website is robotics-transformer-x.github.io. Abigail O'Neill, Abhiram Maddukuri, Abhishek Gupta 0004, Abhishek Padalkar, Abraham Lee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, Ajinkya Jain, Albert Tung, Alex Bewley, Alex Irpan, Alexander Khazatsky, Anant Rai, Anchit Gupta, Andrew E. Wang, Anikait Singh, Animesh Garg, Aniruddha Kembhavi, Annie Xie, Anthony Brohan, Antonin Raffin, Archit Sharma, Arefeh Yavary, Arhan Jain, Ashwin Balakrishna, Ayzaan Wahid, Ben Burgess-Limerick, Bernhard Schölkopf, Blake Wulfe, Brian Ichter, Cewu Lu, Charles Xu 0003, Charlotte Le, Chelsea Finn, Chen Wang 0053, Chenfeng Xu, Cheng Chi 0001, Chenguang Huang, Christine Chan, Christopher Agia, Chuer Pan, Chuyuan Fu, Coline Devin, Danfei Xu, Daniel Morton, Danny Drieß, Daphne Chen, Deepak Pathak, Dhruv Shah, Dieter Büchler, Dinesh Jayaraman, Dmitry Kalashnikov, Dorsa Sadigh, Edward Johns, Ethan Paul Foster, Fangchen Liu, Federico Ceola, Fei Xia 0002, Feiyu Zhao, Freek Stulp, Gaoyue Zhou, Gaurav S. Sukhatme, Gautam Salhotra, Gilbert Feng, Giulio Schiavi, Glen Berseth, Gregory Kahn, Guanzhi Wang, Hao Su 0001, Haoshu Fang, Henghui Bao, Heni Ben Amor, Henrik I. Christensen, Hiroki Furuta, Homer Walke, Hongjie Fang, Huy Ha, Igor Mordatch, Ilija Radosavovic, Isabel Leal, Jacky Liang, Jad Abou-Chakra, Jaehyung Kim 0001, Jaimyn Drake, Jan Peters 0001, Jan Schneider 0007, Jasmine Hsu, Jeannette Bohg, Jeffrey T. Bingham, Jensen Gao, Jiaheng Hu, Jiajun Wu 0001, Jiankai Sun, Jianlan Luo, Jiayuan Gu, Jie Tan 0001, Jihoon Oh, Jimmy Wu, Jingpei Lu, Jitendra Malik, João Silvério, Joey Hejna, Jonathan Booher, Jonathan Tompson, Jonathan Yang, Jordi Salvador, Joseph J. Lim, Junhyek Han, Kanishka Rao, Karl Pertsch, Karol Hausman, Keegan Go, Keerthana Gopalakrishnan, Kenneth Y. Goldberg, Kendra Byrne, Kenneth Oslund, Kento Kawaharazuka, Kevin Black, Kevin Zhang 0002, Kiana Ehsani, Kiran Lekkala, Kirsty Ellis, Krishan Rana, Krishnan Srinivasan, Kuan Fang, Kunal Pratap Singh, Kuo-Hao Zeng, Kyle Hatch, Kyle Hsu, Laurent Itti, Yunliang Chen 0001, Lerrel Pinto, Li Fei-Fei 0001, Liam Tan, Linxi Fan, Lionel Ott, Lisa Lee, Luca Weihs, Magnum Chen, Marion Lepert, Marius Memmel, Masayoshi Tomizuka, Masha Itkina, Mateo Guaman Castro, Max Spero, Maximilian Du, Michael Ahn, Michael C. Yip, Mingtong Zhang 0003, Mingyu Ding, Minho Heo, Mohan Kumar Srirama, Mohit Sharma 0001, Moo Jin Kim, Naoaki Kanazawa, Nicklas Hansen 0001, Nicolas Heess, Nikhil J. Joshi, Niko Sünderhauf, Norman Di Palo, Nur Muhammad Shafiullah, Oier Mees, Oliver Kroemer, Osbert Bastani, Pannag R. Sanketi, Patrick Tree Miller, Patrick Yin, Paul Wohlhart, Peng Xu 0010, Peter David Fagan, Peter Mitrano, Pierre Sermanet, Pieter Abbeel, Priya Sundaresan, Qiuyu Chen, Rafael Rafailov, Ria Doshi, Roberto Martin Martin, Rohan Baijal, Rosario Scalise, Rose Hendrix, Roy Lin, Runjia Qian, Russell Mendonca, Rutav Shah, Ryan Hoque, Ryan Julian, Samuel Bustamante-Gomez, Sean Kirmani, Sergey Levine, Sherry Moore, Shikhar Bahl, Shivin Dass, Shubham D. Sonawani, Shuran Song, Sichun Xu, Siddhant Haldar, Siddharth Karamcheti, Simeon Adebola, Simon Guist, Soroush Nasiriany, Stefan Schaal, Stefan Welker, Stephen Tian, Subramanian Ramamoorthy, Sudeep Dasari, Suneel Belkhale, Sungjae Park, Suraj Nair 0003, Suvir Mirchandani, Takayuki Osa, Tanmay Gupta, Tatsuya Harada, Tatsuya Matsushima, Ted Xiao, Thomas Kollar, Tianhe Yu, Tianli Ding, Todor Davchev, Tony Z. Zhao, Travis Armstrong, Trevor Darrell, Trinity Chung, Vidhi Jain, Vincent Vanhoucke, Wolfram Burgard, Xiaolong Wang 0004, Xinghao Zhu, Xinyang Geng, Liangwei Xu, Yecheng Jason Ma 0001, Yejin Kim 0003, Yevgen Chebotar, Yilin Wu 0003, Yonatan Bisk, Yoonyoung Cho, Youngwoon Lee, Yuchen Cui, Yueh-Hua Wu, Yujin Tang, Yuke Zhu, Yunchu Zhang, Yunfan Jiang 0001, Yunshuang Li, Yunzhu Li, Yusuke Iwasawa, Yutaka Matsuo, Zehan Ma, Zichen Jeff Cui, Zichen Zhang 0016, Zipeng Lin	ICRA	232
2024	Multi-Modal Representation Learning with Tactile Data abstract Advancements in embodied language models like PALM-E and RT-2 have significantly enhanced language-conditioned robotic manipulation. However, these advances remain predominantly focused on vision and language, often overlooking the pivotal role of tactile feedback which is advantageous in contact-rich interactions. Our research introduces a novel approach that synergizes tactile information with vision and language. We present the Multi-Modal Wand (MMWand) dataset enriched with linguistic descriptions and tactile data. By integrating tactile feedback, we aim to bridge the divide between human linguistic understanding and robotic sensory interpretation. Our multi-modal representation model is trained on these datasets by employing the multi-modal embedding alignment principle from ImageBind which has shown promising results, emphasizing the potential of tactile data in robotic applications. The validation of our approach in downstream robotics tasks, such as texture-based object classification, cross-modality retrieval, and the dense reward function for visuomotor control, attests to its effectiveness. Our contributions underscore the importance of tactile feedback in multi-modal robotic learning and its potential to enhance robotic tasks. The MMWand dataset is publicly available at https://hyung-gun.me/mmwand/. Hyung-Gun Chi, Jose A. Barreiros, Jean Mercat, Karthik Ramani, Thomas Kollar	IROS	5
2024	Language-Embedded Gaussian Splats (LEGS): Incrementally Building Room-Scale Representations with a Mobile Robot abstract Building semantic 3D maps is valuable for searching for objects of interest in offices, warehouses, stores, and homes. We present a mapping system that incrementally builds a Language-Embedded Gaussian Splat (LEGS): a detailed 3D scene representation that encodes both appearance and semantics in a unified representation. LEGS is trained online as a robot traverses its environment to enable localization of open-vocabulary object queries. We evaluate LEGS on 4 room-scale scenes where we query for objects in the scene to assess how LEGS can capture semantic meaning. We compare LEGS to LERF [1] and find that while both systems have comparable object query success rates, LEGS trains over 3.5x faster than LERF. Results suggest that a multi-camera setup and incremental bundle adjustment can boost visual reconstruction quality in constrained robot trajectories, and suggest LEGS can localize open-vocabulary and long-tail object queries with up to 66% accuracy. See project website at: berkeleyautomation.github.io/LEGS Justin Yu, Kush Hari, Kishore Srinivas, Karim El-Refai, Adam Rashid, Chung Min Kim, Justin Kerr, Richard Cheng, Muhammad Zubair Irshad, Ashwin Balakrishna, Thomas Kollar, Kenneth Y. Goldberg	IROS	11
2024	MANIP: A Modular Architecture for Integrating Interactive Perception for Robot Manipulation abstract We propose a modular systems architecture, MANIP, that can facilitate the design and development of robot manipulation systems by systematically combining learned subpolicies with well-established procedural algorithmic primitives such as Inverse Kinematics, Kalman Filters, RANSAC outlier rejection, PID modules, etc. (aka "Good Old Fashioned Engineering (GOFE)"). The MANIP architecture grew from our lab’s experience developing robot systems for folding clothes, routing cables, and untangling knots. To address failure modes, MANIP can facilitate inclusion of "interactive perception" subpolicies that execute robot actions to modify system state to bring the system into alignment with the training distribution and / or to disambiguate system state when system state confidence is low. We demonstrate how MANIP can be applied with 3 case studies and then describe a detailed case study in cable tracing with experiments that suggest MANIP can improve performance by up to 88%. Code and details are available at: https://berkeleyautomation.github.io/MANIP/ Justin Yu, Tara Sadjadpour, Abigail O'Neill, Mehdi Khfifi, Yunliang Chen 0001, Richard Cheng, Muhammad Zubair Irshad, Ashwin Balakrishna, Thomas Kollar, Kenneth Y. Goldberg	IROS	9
2024	DataComp-LM: In search of the next generation of training sets for language models abstract We introduce DataComp for Language Models, a testbed for controlled dataset experiments with the goal of improving language models.As part of DCLM, we provide a standardized corpus of 240T tokens extracted from Common Crawl, effective pretraining recipes based on the OpenLM framework, and a broad suite of 53 downstream evaluations.Participants in the DCLM benchmark can experiment with data curation strategies such as deduplication, filtering, and data mixing atmodel scales ranging from 412M to 7B parameters.As a baseline for DCLM, we conduct extensive experiments and find that model-based filtering is key to assembling a high-quality training set.The resulting dataset, DCLM-Baseline, enables training a 7B parameter language model from scratch to 63% 5-shot accuracy on MMLU with 2T training tokens.Compared to MAP-Neo, the previous state-of-the-art in open-data language models, DCLM-Baseline represents a 6 percentage point improvement on MMLU while being trained with half the compute.Our results highlight the importance of dataset design for training language models and offer a starting point for further research on data curation. We release the \dclm benchmark, framework, models, and datasets at https://www.datacomp.ai/dclm/ Jeffrey Li, Alex Fang, Georgios Smyrnis, Maor Ivgi, Matt Jordan, Samir Yitzhak Gadre, Hritik Bansal, Etash Kumar Guha, Sedrick Keh, Kushal Arora, Niklas Muennighoff, Reinhard Heckel, Jean Mercat, Mayee F. Chen, Suchin Gururangan, Mitchell Wortsman, Alon Albalak, Yonatan Bitton, Marianna Nezhurina, Amro Abbas, Cheng-Yu Hsieh, Dhruba Ghosh, Josh Gardner 0001, Maciej Kilian, Hanlin Zhang 0002, Rulin Shao, Sarah M. Pratt, Sunny Sanyal, Gabriel Ilharco, Giannis Daras, Kalyani Marathe, Aaron Gokaslan, Jieyu Zhang 0001, Khyathi Raghavi Chandu, Igor Vasiljevic, Sham M. Kakade, Shuran Song, Sujay Sanghavi, Fartash Faghri, Sewoong Oh, Luke Zettlemoyer, Kyle Lo, Alaaeldin El-Nouby, Hadi Pouransari, Alexander Toshev, Stephanie Wang, Dirk Groeneveld, Luca Soldaini, Pang Wei Koh, Jenia Jitsev, Thomas Kollar, Alexandros G. Dimakis, Yair Carmon, Achal Dave, Ludwig Schmidt, Vaishaal Shankar	NeurIPS	54
2024	A Critical Evaluation of AI Feedback for Aligning Large Language Models abstract Learning from AI feedback (LAIF) is a popular paradigm for improving the instruction-following abilities of powerful pre-trained language models. LAIF first performs supervised fine-tuning (SFT) using demonstrations from a teacher model and then further fine-tunes the model with reinforcement learning (RL) or direct preference optimization (DPO), using feedback from a critic model. While recent popular open-source models have demonstrated substantial improvements in performance from the RL step, in this paper we question whether the complexity of this RL step is truly warranted for AI feedback. We show that the improvements of the RL step are virtually entirely due to the widespread practice of using a weaker teacher model (e.g. GPT-3.5) for SFT data collection than the critic (e.g., GPT-4) used for AI feedback generation. Specifically, we show that simple supervised fine-tuning with GPT-4 as the teacher outperforms existing LAIF pipelines. More generally, we find that the gains from LAIF vary substantially across base model families, test-time evaluation protocols, and critic models. Finally, we provide a mechanistic explanation for when SFT may outperform the full two-step LAIF pipeline as well as suggestions for making LAIF maximally useful in practice. Archit Sharma, Sedrick Keh, Eric Mitchell, Chelsea Finn, Kushal Arora, Thomas Kollar	NeurIPS	6
2023	CARTO: Category and Joint Agnostic Reconstruction of ARTiculated Objects abstract We present CARTO, a novel approach for reconstructing multiple articulated objects from a single stereo RGB observation. We use implicit object-centric representations and learn a single geometry and articulation decoder for multiple object categories. Despite training on multiple categories, our decoder achieves a comparable reconstruction accuracy to methods that train bespoke decoders separately for each category. Combined with our stereo image encoder we infer the 3D shape, 6D pose, size, joint type, and the joint state of multiple unknown objects in a single forward pass. Our method achieves a 20.4% absolute improvement in mAP 3D IOU50 for novel instances when compared to a two-stage pipeline. Inference time is fast and can run on a NVIDIA TITAN XP GPU at 1 HZ for eight or less objects present. While only trained on simulated data, CARTO transfers to real-world object instances. Code and evaluation data is available at: carto. cs. uni - freiburg. de Nick Heppert, Muhammad Zubair Irshad, Sergey Zakharov, Katherine Liu, Rares Ambrus, Jeannette Bohg, Abhinav Valada, Thomas Kollar	CVPR	8
2023	NeO 360: Neural Fields for Sparse View Synthesis of Outdoor Scenes abstract Recent implicit neural representations have shown great results for novel view synthesis. However, existing methods require expensive per-scene optimization from many views hence limiting their application to real-world unbounded urban settings where the objects of interest or backgrounds are observed from very few views. To mitigate this challenge, we introduce a new approach called NeO 360, Neural fields for sparse view synthesis of outdoor scenes. NeO 360 is a generalizable method that reconstructs 360° scenes from a single or a few posed RGB images. The essence of our approach is in capturing the distribution of complex real-world outdoor 3D scenes and using a hybrid image-conditional triplanar representation that can be queried from any world point. Our representation combines the best of both voxel-based and bird’s-eye-view (BEV) representations and is more effective and expressive than each. NeO 360’s representation allows us to learn from a large collection of unbounded 3D scenes while offering generalizability to new views and novel scenes from as few as a single image during inference. We demonstrate our approach on the pro posed challenging 360° unbounded dataset, called NeRDS 360, and show that NeO 360 outperforms state-of-the-art generalizable methods for novel view synthesis while also offering editing and composition capabilities. Project page: zubair-irshad.github.io/projects/neo360.html Muhammad Zubair Irshad, Sergey Zakharov, Katherine Liu, Vitor Campagnolo Guizilini, Thomas Kollar, Adrien Gaidon, Zsolt Kira, Rares Ambrus	ICCV	5
2023	AutoBag: Learning to Open Plastic Bags and Insert Objects abstract Thin plastic bags are ubiquitous in retail stores, healthcare, food handling, recycling, homes, and school lunchrooms. They are challenging both for perception (due to specularities and occlusions) and for manipulation (due to the dynamics of their 3D deformable structure). We formulate the task of “bagging:” manipulating common plastic shopping bags with two handles from an unstructured initial state to an open state where at least one solid object can be inserted into the bag and lifted for transport. We propose a self-supervised learning framework where a dual-arm robot learns to recognize the handles and rim of plastic bags using UV-fluorescent markings; at execution time, the robot does not use UV markings or UV light. We propose the AutoBag algorithm, where the robot uses the learned perception model to open a plastic bag through iterative manipulation. We present novel metrics to evaluate the quality of a bag state and new motion primitives for reorienting and opening bags based on visual observations. In physical experiments, a YuMi robot using AutoBag is able to open bags and achieve a success rate of 16/30 for inserting at least one item across a variety of initial bag configurations. Supplementary material is available at https://sites.google.com/view/autobag. Yunliang Chen 0001, Baiyu Shi, Daniel Seita, Richard Cheng, Thomas Kollar, David Held, Kenneth Y. Goldberg	ICRA	5
2023	SGTM 2.0: Autonomously Untangling Long Cables using Interactive Perception abstract Cables are commonplace in homes, hospitals, and industrial warehouses and are prone to tangling. This paper extends prior work on autonomously untangling long cables by introducing novel uncertainty quantification metrics and actions that interact with the cable to reduce perception uncertainty. We present Sliding and Grasping for Tangle Manipulation 2.0 (SGTM 2.0), a system that autonomously untangles cables approximately 3 meters in length with a bilateral robot using estimates of uncertainty at each step to inform actions. By interactively reducing uncertainty, SGTM 2.0 significantly reduces run-time. Physical experiments with 84 trials suggest that SGTM$2.0$can achieve 83% untangling success on cables with 1 or 2 overhand and figure-8 knots, and 70% termination detection success across these configurations, outperforming SGTM 1.0 by 43% in untangling accuracy and 200% in completion time. Supplementary material, visualizations, and videos can be found at sites.google.com/view/sgtm2. Kaushik Shivakumar, Vainavi Viswanath, Anrui Gu, Yahav Avigal, Justin Kerr, Jeffrey Ichnowski, Richard Cheng, Thomas Kollar, Kenneth Y. Goldberg	ICRA	8
2023	Bagging by Learning to Singulate Layers Using Interactive Perception abstract Many fabric handling and 2D deformable material tasks in homes and industries require singulating layers of material such as opening a bag or arranging garments for sewing. In contrast to methods requiring specialized sensing or end effectors, we use only visual observations with ordinary parallel jaw grippers. We propose SLIP: Singulating Layers using Interactive Perception, and apply SLIP to the task of autonomous bagging. We develop SLIP-Bagging, a bagging algorithm that manipulates a plastic or fabric bag from an unstructured state and uses SLIP to grasp the top layer of the bag to open it for object insertion. In physical experiments, a YuMi robot achieves a success rate of 67% to 81% across bags of a variety of materials, shapes, and sizes, significantly improving in success rate and generality over prior work. Experiments also suggest that SLIP can be applied to tasks such as singulating layers of folded cloth and garments. Supplementary material is available at https://sites.google.com/view/slip-bagging/. Yunliang Chen 0001, Baiyu Shi, Roy Lin, Daniel Seita, Ayah Ahmad, Richard Cheng, Thomas Kollar, David Held, Kenneth Y. Goldberg	IROS	7
2022	ShAPO: Implicit Representations for Multi-object Shape, Appearance, and Pose Optimization Muhammad Zubair Irshad, Sergey Zakharov, Rares Ambrus, Thomas Kollar, Zsolt Kira, Adrien Gaidon	ECCV (2)	4
2022	CenterSnap: Single-Shot Multi-Object 3D Shape Reconstruction and Categorical 6D Pose and Size Estimation abstract This paper studies the complex task of simultaneous multi-object 3D reconstruction, 6D pose and size estimation from a single-view RGB-D observation. In contrast to instance- level pose estimation, we focus on a more challenging problem where CAD models are not available at inference time. Existing approaches mainly follow a complex multi-stage pipeline which first localizes and detects each object instance in the image and then regresses to either their 3D meshes or 6D poses. These approaches suffer from high-computational cost and low performance in complex multi-object scenarios, where occlusions can be present. Hence, we present a simple one- stage approach to predict both the 3D shape and estimate the 6D pose and size jointly in a bounding-box free manner. In particular, our method treats object instances as spatial centers where each center denotes the complete shape of an object along with its 6D pose and size. Through this per- pixel representation, our approach can reconstruct in real- time (40 FPS) multiple novel object instances and predict their 6D pose and sizes in a single-forward pass. Through extensive experiments, we demonstrate that our approach significantly outperforms all shape completion and categorical 6D pose and size estimation baselines on multi-object ShapeNet and NOCS datasets respectively with a 12.6% absolute improvement in mAP for 6D pose for novel real-world object instances. Muhammad Zubair Irshad, Thomas Kollar, Michael Laskey, Kevin Stone, Zsolt Kira	ICRA	2
2022	Efficiently Learning Single-Arm Fling Motions to Smooth Garments Yunliang Chen 0001, Ellen R. Novoseller, Daniel Seita, Jeffrey Ichnowski, Michael Laskey, Richard Cheng, Thomas Kollar, Kenneth Y. Goldberg	ISRR	8
2020	A Mobile Manipulation System for One-Shot Teaching of Complex Tasks in Homes abstract We describe a mobile manipulation hardware and software system capable of autonomously performing complex human-level tasks in real homes, after being taught the task with a single demonstration from a person in virtual reality. This is enabled by a highly capable mobile manipulation robot, whole-body task space hybrid position/force control, teaching of parameterized primitives linked to a robust learned dense visual embeddings representation of the scene, and a task graph of the taught behaviors. We demonstrate the robustness of the approach by presenting results for performing a variety of tasks, under different environmental conditions, in multiple real homes. Our approach achieves 85% overall success rate on three tasks that consist of an average of 45 behaviors each. The video is available at: https://youtu.be/HSyAGMGikLk. Max Bajracharya, James Borders, Daniel M. Helmick, Thomas Kollar, Michael Laskey, John Leichty, Jeremy Ma, Umashankar Nagarajan, Akiyoshi Ochiai, Josh Petersen, Krishna Shankar, Kevin Stone, Yutaka Takaoka	ICRA	4
2018	Multi-Task Learning For Parsing The Alexa Meaning Representation Language abstract The Alexa Meaning Representation Language (AMRL) is a compositional graph-based semantic representation that includes fine-grained types, properties, actions, and roles and can represent a wide variety of spoken language. AMRL increases the ability of virtual assistants to represent more complex requests, including logical and conditional statements as well as ones with nested clauses. Due to this representational capacity, the acquisition of large scale data resources is challenging, which limits the accuracy of resulting models. This paper has two primary contributions. First, we develop a linearization of AMRL graphs along with a deep multi-task model that predicts fine-grained types, properties, and intents. Second, we show how to jointly train a model that predicts an existing representation for spoken language understanding (SLU) along with the linearized AMRL parse. The resulting model, which leverages learned embeddings from both tasks, is able to predict the AMRL representation more accurately than other approaches, decreasing the error rates in the full parse by 3.56% absolute and reducing the amount of natively annotated data needed to train accurate parsing models. Vittorio Perera, Tagyoung Chung, Thomas Kollar, Emma Strubell	AAAI	3
2013	Imitation learning for natural language direction following through unknown environments abstract The use of spoken instructions in human-robot teams holds the promise of enabling untrained users to effectively control complex robotic systems in a natural and intuitive way. Providing robots with the capability to understand natural language directions would enable effortless coordination in human robot teams that operate in non-specialized unknown environments. However, natural language direction following through unknown environments requires understanding the meaning of language, using a partial semantic world model to generate actions in the world, and reasoning about the environment and landmarks that have not yet been detected. We address the problem of robots following natural language directions through complex unknown environments. By exploiting the structure of spatial language, we can frame direction following as a problem of sequential decision making under uncertainty. We learn a policy which predicts a sequence of actions that follow the directions by exploring the environment and discovering landmarks, backtracking when necessary, and explicitly declaring when it has reached the destination. We use imitation learning to train the policy, using demonstrations of people following directions. By training explicitly in unknown environments, we can generalize to situations that have not been encountered previously. Felix Duvallet, Thomas Kollar, Anthony Stentz	ICRA	2
2013	Learning environmental knowledge from task-based human-robot dialog abstract This paper presents an approach for learning environmental knowledge from task-based human-robot dialog. Previous approaches to dialog use domain knowledge to constrain the types of language people are likely to use. In contrast, by introducing a joint probabilistic model over speech, the resulting semantic parse and the mapping from each element of the parse to a physical entity in the building (e.g., grounding), our approach is flexible to the ways that untrained people interact with robots, is robust to speech to text errors and is able to learn referring expressions for physical locations in a map (e.g., to create a semantic map). Our approach has been evaluated by having untrained people interact with a service robot. Starting with an empty semantic map, our approach is able ask 50% fewer questions than a baseline approach, thereby enabling more effective and intuitive human robot dialog. Thomas Kollar, Vittorio Perera, Daniele Nardi, Manuela M. Veloso	ICRA	1
2013	Jointly Learning to Parse and Perceive: Connecting Natural Language to the Physical World abstract This paper introduces Logical Semantics with Perception (LSP), a model for grounded language acquisition that learns to map natural language statements to their referents in a physical environment. For example, given an image, LSP can map the statement “blue mug on the table” to the set of image segments showing blue mugs on tables. LSP learns physical representations for both categorical (“blue,” “mug”) and relational (“on”) language, and also learns to compose these representations to produce the referents of entire statements. We further introduce a weakly supervised training procedure that estimates LSP’s parameters using annotated referents for entire statements, without annotated referents for individual words or the parse structure of the statement. We perform experiments on two applications: scene understanding and geographical question answering. We find that LSP outperforms existing, less expressive models that cannot represent relational language. We further find that weakly supervised training is competitive with fully supervised training while requiring significantly less annotation effort. Jayant Krishnamurthy, Thomas Kollar	Trans. Assoc. Comput. Linguistics	2
2013	Clarifying commands with information-theoretic human-robot dialog abstract Our goal is to improve the efficiency and effectiveness of natural language communication between humans and robots. Human language is frequently ambiguous, and a robot's limited sensing makes complete understanding of a statement even more difficult. To address these challenges, we describe an approach for enabling a robot to engage in clarifying dialog with a human partner, just as a human might do in a similar situation. Given an unconstrained command from a human operator, the robot asks one or more questions and receives natural language answers from the human. We apply an information-theoretic approach to choosing questions for the robot to ask. Specifically, we choose the type and subject of questions in order to maximize the reduction in Shannon entropy of the robot's mapping between language and entities in the world. Within the framework of the G3 graphical model, we derive a method to estimate this entropy reduction, choose the optimal question to ask, and merge the information gained from the human operator's answer. We demonstrate that this improves the accuracy of command understanding over prior work while asking fewer questions as compared to baseline question-selection strategies. Robin Deits, Stefanie Tellex, Pratiksha Thaker, Dimitar Simeonov, Thomas Kollar, Nicholas Roy	J. Hum. Robot Interact.	5
2012	Using the Web to Interactively Learn to Find Objects abstract In order for robots to intelligently perform tasks with humans, they must be able to access a broad set of background knowledge about the environments in which they operate. Unlike other approaches, which tend to manually define the knowledge of the robot, our approach enables robots to actively query the World Wide Web (WWW) to learn background knowledge about the physical environment. We show that our approach is able to search the Web to infer the probability that an object, such as a "coffee,'' can be found in a location, such as a "kitchen.'' Our approach, called ObjectEval, is able to dynamically instantiate a utility function using this probability, enabling robots to find arbitrary objects in indoor environments. Our experimental results show that the interactive version of ObjectEval visits 28% fewer locations than the version trained offline and 71% fewer locations than a baseline approach which uses no background knowledge. Mehdi Samadi, Thomas Kollar, Manuela M. Veloso	AAAI	2
2012	CoBots: Collaborative robots servicing multi-floor buildings abstract In this video we briefly illustrate the progress and contributions made with our mobile, indoor, service robots CoBots (Collaborative Robots), since their creation in 2009. Many researchers, present authors included, aim for autonomous mobile robots that robustly perform service tasks for humans in our indoor environments. The efforts towards this goal have been numerous and successful, and we build upon them. However, there are clearly many research challenges remaining until we can experience intelligent mobile robots that are fully functional and capable in our human environments. Manuela M. Veloso, Joydeep Biswas, Brian Coltin, Stephanie Rosenthal, Thomas Kollar, Çetin Meriçli, Mehdi Samadi, Susana Brandão, Rodrigo M. M. Ventura	IROS	5
2011	Understanding Natural Language Commands for Robotic Navigation and Mobile Manipulation abstract This paper describes a new model for understanding natural language commands given to autonomous systems that perform navigation and mobile manipulation in semi-structured environments. Previous approaches have used models with fixed structure to infer the likelihood of a sequence of actions given the environment and the command. In contrast, our framework, called Generalized Grounding Graphs, dynamically instantiates a probabilistic graphical model for a particular natural language command according to the command's hierarchical and compositional semantic structure. Our system performs inference in the model to successfully find and execute plans corresponding to natural language commands such as "Put the tire pallet on the truck." The model is trained using a corpus of commands collected using crowdsourcing. We pair each command with robot actions and use the corpus to learn the parameters of the model. We evaluate the robot's performance by inferring plans from natural language commands, executing each plan in a realistic robot simulator, and asking users to evaluate the system's performance. We demonstrate that our system can successfully follow many natural language commands from the corpus. Stefanie Tellex, Thomas Kollar, Steven Dickerson, Matthew R. Walter, Ashis Gopal Banerjee, Seth J. Teller, Nicholas Roy	AAAI	2
2011	HRI pioneers workshop 2011 abstract The 2011 HRI Pioneers Workshop will be conducted in conjunction with the 2011 ACM/IEEE International Conference on Human-Robot Interaction (HRI). The 2011 HRI Pioneers Workshop will provide a forum for graduate students and postdocs to learn about the current state of HRI, to present their work and to network with one another and with select senior researchers in a setting that is less formal and more interactive than the main conference. Workshop participants will discuss important issues and open challenges in the field, encouraging the formation of collaborative relationships across disciplines and geographic boundaries. Thomas Kollar, Astrid Weiss, Jason Monast, Anja Austermann, David Lu, Elena Gribovskaya, Chandan Datta, Richard Kelley, Hirotaka Osawa, Lanny Lin	HRI	1
2011	Following and interpreting narrated guided tours abstract We describe a robotic tour-taking capability enabling a robot to acquire local knowledge of a human-occupied environment. A tour-taking robot autonomously follows a human guide through an environment, interpreting the guide's spoken utterances and the shared spatiotemporal context in order to acquire a spatially segmented and semantically labeled metrical-topological representation of the environment. The described tour-taking capability enables scalable deployment of mobile robots into human-occupied environments, and natural human-robot interaction for commanded mobility. Our primary contributions are an efficient, socially acceptable autonomous tour-following behavior and a tour interpretation algorithm that partitions a map into spaces labeled according to the guide's utterances. The tour-taking behavior is demonstrated in a multi-floor office building and evaluated by assessing the comfort of the tour guides, and by comparing the robot's map partitions to those produced by humans. Sachithra Hemachandra, Thomas Kollar, Nicholas Roy, Seth J. Teller	ICRA	2
2010	Toward understanding natural language directions abstract Speaking using unconstrained natural language is an intuitive and flexible way for humans to interact with robots. Understanding this kind of linguistic input is challenging because diverse words and phrases must be mapped into structures that the robot can understand, and elements in those structures must be grounded in an uncertain environment. We present a system that follows natural language directions by extracting a sequence of spatial description clauses from the linguistic input and then infers the most probable path through the environment given only information about the environmental geometry and detected visible objects. We use a probabilistic graphical model that factors into three key components. The first component grounds landmark phrases such as "the computers" in the perceptual frame of the robot by exploiting co-occurrence statistics from a database of tagged images such as Flickr. Second, a spatial reasoning component judges how well spatial relations such as "past the computers" describe a path. Finally, verb phrases such as "turn right" are modeled according to the amount of change in orientation in the path. Our system follows 60% of the directions in our corpus to within 15 meters of the true destination, significantly outperforming other approaches. Thomas Kollar, Stefanie Tellex, Deb Roy, Nicholas Roy	HRI	1
2010	Indoor scene recognition through object detection abstract Scene recognition is a highly valuable perceptual ability for an indoor mobile robot, however, current approaches for scene recognition present a significant drop in performance for the case of indoor scenes. We believe that this can be explained by the high appearance variability of indoor environments. This stresses the need to include high-level semantic information in the recognition process. In this work we propose a new approach for indoor scene recognition based on a generative probabilistic hierarchical model that uses common objects as an intermediate semantic representation. Under this model, we use object classifiers to associate low-level visual features to objects, and at the same time, we use contextual relations to associate objects to scenes. As a further contribution, we improve the performance of current state-of-the-art category-level object classifiers by including geometrical information obtained from a 3D range sensor that facilitates the implementation of a focus of attention mechanism within a Monte Carlo sampling scheme. We test our approach using real data, showing significant advantages with respect to previous state-of-the-art methods. Pablo Espinace, Thomas Kollar, Alvaro Soto, Nicholas Roy	ICRA	2
2010	Natural language command of an autonomous micro-air vehicle abstract Natural language is a flexible and intuitive modality for conveying directions and commands to a robot but presents a number of computational challenges. Diverse words and phrases must be mapped into structures that the robot can understand, and elements in those structures must be grounded in an uncertain environment. In this paper we present a micro-air vehicle (MAV) capable of following natural language directions through a previously mapped and labeled environment. We extend our previous work in understanding 2D natural language directions to three dimensions, accommodating new verb modifiers such as go up and go down, and commands such as turn around and face the windows. We demonstrate the robot following directions created by a human for another human, and interactively executing commands in the context of surveillance and search and rescue in confined spaces. In an informal study, 71% of the paths computed from directions given by one user terminated within 10 m of the desired destination. Albert S. Huang, Stefanie Tellex, Abraham Bachrach, Thomas Kollar, Deb Roy, Nicholas Roy	IROS	4
2009	Utilizing object-object and object-scene context when planning to find things abstract In this paper, our goal is to search for a novel object, where we have a prior map of the environment and knowledge of some of the objects in it, but no information about the location of the specific novel object. We develop a probabilistic model over possible object locations that utilizes object-object and object-scene context. This model can be queried for any of over 25,000 naturally occurring objects in the world and is trained from labeled data acquired from the captions of photos on the Flickr Website. We show that these simple models based on object co-occurrences perform surprisingly well at localizing arbitrary objects in an office setting. In addition, we show how to compute paths that minimize the expected distance to the query object and show that this approach performs better than a greedy approach. Finally, we give preliminary results for grounding our approach in object classifiers. Thomas Kollar, Nicholas Roy	ICRA	1
2009	Where to go: Interpreting natural directions using global inference abstract An important component of human-robot interaction is that people need to be able to instruct robots to move to other locations using naturally given directions. When giving directions, people often make mistakes such as labelling errors (e.g., left vs. right) and errors of omission (skipping important decision points in a sequence). Furthermore, people often use multiple levels of granularity in specifying directions, referring to locations using single object landmarks, multiple landmarks in a given location, or identifying large regions as a single location. The challenge is to identify the correct path to a destination from a sequence of noisy, possibly erroneous directions. In our work we cast this problem as probabilistic inference: given a set of directions, an agent should automatically find the path with the geometry and physical appearance to maximize the likelihood of those directions. We use a specific variant of a Markov Random Field (MRF) to represent our model, and gather multi-granularity representation information using existing large tagged datasets. On a dataset of route directions collected in a large third floor university building, we found that our algorithm correctly inferred the true final destination in 47 out of the 55 cases successfully followed by humans volunteers. These results suggest that our algorithm is performing well relative to human users. In the future this work will be included in a broader system for autonomously constructing environmental representations that support natural human-robot interaction for direction giving. Emma Brunskill, Thomas Kollar, Nicholas Roy	ICRA	3
2008	Efficient Optimization of Information-Theoretic Exploration in SLAM Thomas Kollar, Nicholas Roy	AAAI	1
2007	Topological mapping using spectral clustering and classification abstract In this work we present an online method for generating topological maps from raw sensor information. We first describe an algorithm to automatically decompose a map into submap segments using a graph partitioning technique known as spectral clustering. We then describe how to train a classifier to recognize graph submaps from laser signatures using the AdaBoost machine learning algorithm. We demonstrate that the we can perform topological mapping by incrementally segmenting the world as the robot moves through its environment, and we can close the loop when the learned classifier recognizes that the robot has returned to a previously visited location. Emma Brunskill, Thomas Kollar, Nicholas Roy	IROS	2
2007	Collision detection in legged locomotion using supervised learning abstract We propose a fast approach for detecting collision- free swing-foot trajectories for legged locomotion over extreme terrains. Instead of simulating the swing trajectories and checking for collisions along them, our approach uses machine learning techniques to predict whether a swing trajectory is collision-free. Using a set of local terrain features, we apply supervised learning to train a classifier to predict collisions. Both in simulation and on a real quadruped platform, our results show that our classifiers can improve the accuracy of collision detection compared to a real-time geometric approach without significantly increasing the computation time. Finale Doshi-Velez, Emma Brunskill, Alexander C. Shkolnik, Thomas Kollar, Khashayar Rohanimanesh, Russ Tedrake, Nicholas Roy	IROS	4
2006	Using Reinforcement Learning to Improve Exploration Trajectories for Error Minimization abstract The mapping and localization problems have received considerable attention in robotics recently. The exploration problem that drives mapping has started to generate similar attention, as the ease of construction and quality of map is strongly dependent on the strategy used to acquire sensor data for the map. Most exploration strategies concentrate on selecting the next best measurement to take, trading off information gathering for regular relocalization. What has not been studied so far is the effect the robot controller has on the map quality while executing exploration plans. Certain kinds of robot motion (e.g, sharp turns) are hard to estimate correctly, and increase the likelihood of errors in the mapping process. We show how reinforcement learning can be used to generate good motion control while executing a simple information gathering exploration strategy. We show that the learned policy reduces the overall map uncertainty by reducing the amount of uncertainty generated by robot motion Thomas Kollar, Nicholas Roy	ICRA	1