Search Results for author: David Held

Found 39 papers, 20 papers with code

Self-Supervised Point Cloud Completion via Inpainting

no code implementations21 Nov 2021 Himangi Mittal, Brian Okorn, Arpit Jangid, David Held

The aim of this work is to learn to complete these partial point clouds, giving us a full understanding of the object's geometry using only partial observations.

Point Cloud Completion

FabricFlowNet: Bimanual Cloth Manipulation with a Flow-based Policy

no code implementations10 Nov 2021 Thomas Weng, Sujay Bajracharya, YuFei Wang, Khush Agrawal, David Held

We introduce FabricFlowNet (FFN), a cloth manipulation policy that leverages flow as both an input and as an action representation to improve performance.

Motion Estimation Optical Flow Estimation

Active Safety Envelopes using Light Curtains with Probabilistic Guarantees

no code implementations8 Jul 2021 Siddharth Ancha, Gaurav Pathak, Srinivasa G. Narasimhan, David Held

We use light curtains to estimate the safety envelope of a scene: a hypothetical surface that separates the robot from all obstacles.

Exploiting & Refining Depth Distributions With Triangulation Light Curtains

no code implementations CVPR 2021 Yaadhav Raaj, Siddharth Ancha, Robert Tamburo, David Held, Srinivasa G. Narasimhan

Active sensing through the use of Adaptive Depth Sensors is a nascent field, with potential in areas such as Advanced driver-assistance systems (ADAS).

Learning Visible Connectivity Dynamics for Cloth Smoothing

no code implementations21 May 2021 Xingyu Lin, YuFei Wang, Zixuan Huang, David Held

Robotic manipulation of cloth remains challenging for robotics due to the complex dynamics of the cloth, lack of a low-dimensional state representation, and self-occlusions.

ZePHyR: Zero-shot Pose Hypothesis Rating

1 code implementation28 Apr 2021 Brian Okorn, Qiao Gu, Martial Hebert, David Held

We also demonstrate how our system can be used by quickly scanning and building a model of a novel object, which can immediately be used by our method for pose estimation.

Motion Planning Pose Estimation

Lyapunov Barrier Policy Optimization

1 code implementation16 Mar 2021 Harshit Sikchi, Wenxuan Zhou, David Held

Current RL agents explore the environment without considering these constraints, which can lead to damage to the hardware or even other agents in the environment.

PanoNet3D: Combining Semantic and Geometric Understanding for LiDARPoint Cloud Detection

no code implementations17 Dec 2020 Xia Chen, Jianren Wang, David Held, Martial Hebert

Visual data in autonomous driving perception, such as camera image and LiDAR point cloud, can be interpreted as a mixture of two aspects: semantic feature and geometric structure.

Autonomous Driving Cloud Detection

PLAS: Latent Action Space for Offline Reinforcement Learning

3 code implementations14 Nov 2020 Wenxuan Zhou, Sujay Bajracharya, David Held

The goal of offline reinforcement learning is to learn a policy from a fixed dataset, without further interactions with the environment.

Continuous Control Deformable Object Manipulation

SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation

1 code implementation14 Nov 2020 Xingyu Lin, YuFei Wang, Jake Olkin, David Held

Further, we evaluate a variety of algorithms on these tasks and highlight challenges for reinforcement learning algorithms, including dealing with a state representation that has a high intrinsic dimensionality and is partially observable.

Deformable Object Manipulation OpenAI Gym

ROLL: Visual Self-Supervised Reinforcement Learning with Object Reasoning

1 code implementation13 Nov 2020 YuFei Wang, Gautham Narayan Narasimhan, Xingyu Lin, Brian Okorn, David Held

Current image-based reinforcement learning (RL) algorithms typically operate on the whole image without performing object-level reasoning.

Multi-Goal Reinforcement Learning

Robust Instance Tracking via Uncertainty Flow

no code implementations9 Oct 2020 Jianing Qian, Junyu Nan, Siddharth Ancha, Brian Okorn, David Held

Current state-of-the-art trackers often fail due to distractorsand large object appearance changes.

Optical Flow Estimation

Learning Off-Policy with Online Planning

no code implementations23 Aug 2020 Harshit Sikchi, Wenxuan Zhou, David Held

In this work, we investigate a novel instantiation of H-step lookahead with a learned model and a terminal value function learned by a model-free off-policy algorithm, named Learning Off-Policy with Online Planning (LOOP).

Continuous Control

AB3DMOT: A Baseline for 3D Multi-Object Tracking and New Evaluation Metrics

no code implementations18 Aug 2020 Xinshuo Weng, Jianren Wang, David Held, Kris Kitani

Additionally, 3D MOT datasets such as KITTI evaluate MOT methods in 2D space and standardized 3D MOT evaluation tools are missing for a fair comparison of 3D MOT methods.

3D Multi-Object Tracking Autonomous Driving

Uncertainty-aware Self-supervised 3D Data Association

1 code implementation18 Aug 2020 Jianren Wang, Siddharth Ancha, Yi-Ting Chen, David Held

Instead, we propose leveraging vast unlabeled datasets by self-supervised metric learning of 3D object trackers, with a focus on data association.

Metric Learning Object Detection

Cloth Region Segmentation for Robust Grasp Selection

1 code implementation13 Aug 2020 Jianing Qian, Thomas Weng, Luxin Zhang, Brian Okorn, David Held

Our approach trains a network to segment the edges and corners of a cloth from a depth image, distinguishing such regions from wrinkles or folds.

Robotics

Active Perception using Light Curtains for Autonomous Driving

no code implementations ECCV 2020 Siddharth Ancha, Yaadhav Raaj, Peiyun Hu, Srinivasa G. Narasimhan, David Held

Most real-world 3D sensors such as LiDARs perform fixed scans of the entire environment, while being decoupled from the recognition system that processes the sensor data.

3D Object Recognition Autonomous Driving

Learning Orientation Distributions for Object Pose Estimation

1 code implementation2 Jul 2020 Brian Okorn, Mengyun Xu, Martial Hebert, David Held

Our first method, which regresses from deep learned features to an isotropic Bingham distribution, gives the best performance for orientation distribution estimation for non-symmetric objects.

Pose Estimation

Multi-modal Transfer Learning for Grasping Transparent and Specular Objects

no code implementations29 May 2020 Thomas Weng, Amith Pallankize, Yimin Tang, Oliver Kroemer, David Held

State-of-the-art object grasping methods rely on depth sensing to plan robust grasps, but commercially available depth sensors fail to detect transparent and specular objects.

Robotics

Combining Deep Learning and Verification for Precise Object Instance Detection

2 code implementations27 Dec 2019 Siddharth Ancha, Junyu Nan, David Held

For a reliable detection system, if a high confidence detection is made, we would want high certainty that the object has indeed been detected.

Learning to Optimally Segment Point Clouds

no code implementations10 Dec 2019 Peiyun Hu, David Held, Deva Ramanan

We prove that if we score a segmentation by the worst objectness among its individual segments, there is an efficient algorithm that finds the optimal worst-case segmentation among an exponentially large number of candidate segmentations.

Instance Segmentation Semantic Segmentation

What You See is What You Get: Exploiting Visibility for 3D Object Detection

1 code implementation CVPR 2020 Peiyun Hu, Jason Ziglar, David Held, Deva Ramanan

On the NuScenes 3D detection benchmark, we show that, by adding an additional stream for visibility input, we can significantly improve the overall detection accuracy of a state-of-the-art 3D detector.

3D Object Detection Data Augmentation

Adaptive Auxiliary Task Weighting for Reinforcement Learning

1 code implementation NeurIPS 2019 Xingyu Lin, Harjatin Baweja, George Kantor, David Held

Reinforcement learning is known to be sample inefficient, preventing its application to many real-world problems, especially with high dimensional observations like images.

Just Go with the Flow: Self-Supervised Scene Flow Estimation

1 code implementation CVPR 2020 Himangi Mittal, Brian Okorn, David Held

When interacting with highly dynamic environments, scene flow allows autonomous systems to reason about the non-rigid motion of multiple independent objects.

Autonomous Driving Scene Flow Estimation

3D Multi-Object Tracking: A Baseline and New Evaluation Metrics

1 code implementation9 Jul 2019 Xinshuo Weng, Jianren Wang, David Held, Kris Kitani

Additionally, 3D MOT datasets such as KITTI evaluate MOT methods in the 2D space and standardized 3D MOT evaluation tools are missing for a fair comparison of 3D MOT methods.

3D Multi-Object Tracking Autonomous Driving +1

Few-Shot Point Cloud Region Annotation with Human in the Loop

1 code implementation11 Jun 2019 Siddhant Jain, Sowmya Munukutla, David Held

We propose a point cloud annotation framework that employs human-in-loop learning to enable the creation of large point cloud datasets with per-point annotations.

Few-Shot Learning Fine-tuning

Reinforcement Learning without Ground-Truth State

no code implementations20 May 2019 Xingyu Lin, Harjatin Singh Baweja, David Held

However, if this policy is trained with reinforcement learning, then without a state estimator, it is hard to specify a reward function based on high-dimensional observations.

Adaptive Variance for Changing Sparse-Reward Environments

no code implementations15 Mar 2019 Xingyu Lin, Pengsheng Guo, Carlos Florensa, David Held

Robots that are trained to perform a task in a fixed environment often fail when facing unexpected changes to the environment due to a lack of exploration.

Iterative Transformer Network for 3D Point Cloud

1 code implementation27 Nov 2018 Wentao Yuan, David Held, Christoph Mertz, Martial Hebert

Recently, neural networks operating on point clouds have shown superior performance on 3D understanding tasks such as shape classification and part segmentation.

General Classification

PCN: Point Completion Network

4 code implementations2 Aug 2018 Wentao Yuan, Tejas Khot, David Held, Christoph Mertz, Martial Hebert

Shape completion, the problem of estimating the complete geometry of objects from partial observations, lies at the core of many vision and robotics applications.

Point Cloud Completion

Reverse Curriculum Generation for Reinforcement Learning

no code implementations17 Jul 2017 Carlos Florensa, David Held, Markus Wulfmeier, Michael Zhang, Pieter Abbeel

The robot is trained in reverse, gradually learning to reach the goal from a set of start states increasingly far from the goal.

Constrained Policy Optimization

6 code implementations ICML 2017 Joshua Achiam, David Held, Aviv Tamar, Pieter Abbeel

For many applications of reinforcement learning it can be more convenient to specify both a reward function and constraints, rather than trying to design behavior through the reward function.

Automatic Goal Generation for Reinforcement Learning Agents

1 code implementation ICML 2018 Carlos Florensa, David Held, Xinyang Geng, Pieter Abbeel

Instead, we propose a method that allows an agent to automatically discover the range of tasks that it is capable of performing.

Probabilistically Safe Policy Transfer

no code implementations15 May 2017 David Held, Zoe McCarthy, Michael Zhang, Fred Shentu, Pieter Abbeel

Although learning-based methods have great potential for robotics, one concern is that a robot that updates its parameters might cause large amounts of damage before it learns the optimal policy.

Enabling Robots to Communicate their Objectives

no code implementations11 Feb 2017 Sandy H. Huang, David Held, Pieter Abbeel, Anca D. Dragan

We show that certain approximate-inference models lead to the robot generating example behaviors that better enable users to anticipate what it will do in novel situations.

Autonomous Driving

Learning to Track at 100 FPS with Deep Regression Networks

3 code implementations6 Apr 2016 David Held, Sebastian Thrun, Silvio Savarese

We propose a method for offline training of neural networks that can track novel objects at test-time at 100 fps.

Deep Learning for Single-View Instance Recognition

no code implementations29 Jul 2015 David Held, Sebastian Thrun, Silvio Savarese

We show that feedforward neural networks outperform state-of-the-art methods for recognizing objects from novel viewpoints even when trained from just a single image per object.

Cannot find the paper you are looking for? You can Submit a new open access paper.