ICRA 2021 Papers

Learning from Hallucination (LfH) is a recent machine learning paradigm for autonomous navigation, which uses training data collected in completely safe environments and adds numerous imaginary obstacles to make the environment densely constrained, to learn navigation planners that produce feasible navigation even in highly constrained (more dangerous) spaces. However, LfH requires hallucinating the robot perception during deployment to match with the hallucinated training data, which creates a need for sometimes-infeasible prior knowledge and tends to generate very conservative planning. In this work, we propose a new LfH paradigm that does not require runtime hallucination—a feature we call "sober deployment"—and can therefore adapt to more realistic navigation scenarios. This novel Hallucinated Learning and Sober Deployment (HLSD) paradigm is tested in a benchmark testbed of 300 simulated navigation environments with a wide range of difficulty levels, and in the real-world. In most cases, HLSD outperforms both the original LfH method and a classical navigation planner.

The kinematic features of a centaur-type humanoid platform, combined with a powerful actuation, enable the experimentation of a variety of agile and dynamic motions. However, the higher number of degrees-of-freedom and the increased weight of the system, compared to the bipedal and quadrupedal counterparts, pose significant research challenges in terms of computational load and real implementation. To this end, this work presents a control architecture to perform agile actions, conceived for torque-controlled platforms, which decouples for computational purposes offline optimal control planning of lower-body primitives, based on a template kinematic model, and online control of the upper-body motion to maintain balance. Three stabilizing strategies are presented, whose performance is compared in two types of simulated jumps, while experimental validation is performed on a half-squat jump using the CENTAURO robot.

Analyzing surgical workflow is crucial for surgical assistance robots to understand surgeries. With the understanding of the complete surgical workflow, the robots are able to assist the surgeons in intra-operative events, such as by giving a warning when the surgeon is entering specific keys or high-risk phases. Deep learning techniques have recently been widely applied to recognizing surgical workflows. Many of the existing temporal neural network models are limited in their capability to handle long-term dependencies in the data, instead, relying upon the strong performance of the underlying per-frame visual models. We propose a new temporal network structure that leverages task-specific network representation to collect long-term sufficient statistics that are propagated by a sufficient statistics model (SSM). We implement our approach within an LSTM backbone for the task of surgical phase recognition and explore several choices for propagated statistics. We demonstrate superior results over existing and novel state-of-the-art segmentation techniques on two laparoscopic cholecystectomy datasets: the publicly available Cholec80 dataset and MGH100, a novel dataset with more challenging and clinically meaningful segment labels.

It is widely anticipated that Level-44 (L4) autonomous vehicles - vehicles capable of autonomously performing all driving functions under certain conditions - will eventually become available to the public. However, these vehicles are likely to be expensive due to the additional requirements on hardware, making them potentially inaccessible to the average consumer. This paper introduces the concept of Affordable Autonomy through Cooperative Sensing and Planning (AACSP). This concept enables a vehicle equipped with lower-end sensors such as a Level-26 vehicle - vehicle capable of limited lateral and longitudinal control but requiring a driver to stay engaged - to enter into an autonomous mode whenever it is in proximity with a Level-4 vehicle traveling along the same route. The paper describes the system along with the technical components of cooperative planning & state coordination in detail while giving an overview of the cooperative perception and localization modules that are required to implement the AACSP concept. The paper also presents results from simulation and real vehicles.

Vehicle Re-identification (Re-ID) aims to retrieve all instances of query vehicle images present in an image pool. However viewpoint, illumination, and occlusion variations along with subtle differences between two unique images pose a significant challenge towards achieving an effective system. In this paper, we emphasize upon enhancing the performance of visual feature based ReID system by improving feature embedding quality and propose (1) an attention-guided hierarchical feature extractor (HFE) that leverages the structure of a backbone CNN to extract coarse and fine-grained features and (2) to train the proposed network within a hard negative adversarial framework that generates samples exhibiting extreme variations, encouraging the network to extract important distinguishing features across varying scales. To demonstrate the effectiveness of the proposed framework we use VERI-Wild, VRIC and Veri-776 datasets that exhibit extreme intra-class and minute inter-class differences and achieve state-of-the-art (SoTA) performance. Codes related to this paper are publicly available at https://github.com/PS06/VReID.

Adversarial training is an effective method to train deep learning models that are resilient to norm-bounded perturbations, with the cost of nominal performance drop. While adversarial training appears to enhance the robustness and safety of a deep model deployed in open-world decision-critical applications, counterintuitively, it induces undesired behaviors in robot learning settings. In this paper, we show theoretically and experimentally that neural controllers obtained via adversarial training are subjected to three types of defects, namely transient, systematic, and conditional errors. We first generalize adversarial training to a safety-domain optimization scheme allowing for more generic specifications. We then prove that such a learning process tends to cause certain error profiles. We support our theoretical results by a thorough experimental safety analysis in a robot-learning task. Our results suggest that adversarial training is not yet ready for robot learning.

Deep reinforcement learning has made significant progress in robotic manipulation tasks and it works well in the ideal disturbance-free environment. However, in a real-world environment, both internal and external disturbances are inevitable, thus the performance of the trained policy will dramatically drop. To improve the robustness of the policy, we introduce the adversarial training mechanism to the robotic manipulation tasks in this paper, and an adversarial skill learning algorithm based on soft actor-critic (SAC) is proposed for robust manipulation. Extensive experiments are conducted to demonstrate that the learned policy is robust to internal and external disturbances. Additionally, the proposed algorithm is evaluated in both the simulation environment and on the real robotic platform.

Deep Imitation Learning requires a large number of expert demonstrations, which are not always easy to obtain, especially for complex tasks. A way to overcome this shortage of labels is through data augmentation. However, this cannot be easily applied to control tasks due to the sequential nature of the problem. In this work, we introduce a novel augmentation method which preserves the success of the augmented trajectories. To achieve this, we introduce a semi-supervised correction network that aims to correct distorted expert actions. To adequately test the abilities of the correction network, we develop an adversarial data augmented imitation architecture to train an imitation agent using synthetic experts. Additionally, we introduce a metric to measure diversity in trajectory datasets. Experiments show that our data augmentation strategy can improve accuracy and convergence time of adversarial imitation while preserving the diversity between the generated and real trajectories.

Autonomous systems often rely on neural networks to achieve high performance on planning and control problems. Unfortunately, neural networks suffer severely when input images become degraded in ways that are not reflected in the training data. This is particularly problematic for robotic systems like autonomous vehicles (AV) for which reliability is paramount. In this work, we consider robust optimization methods for hardening control systems against image corruptions and other unexpected domain shifts. Recent work on robust optimization for neural nets has been focused largely on combating adversarial attacks. In this work, we borrow ideas from the adversarial training and data augmentation literature to enhance robustness to image corruptions and domain shifts. To this end, we train networks while augmenting image data with a battery of image degradations. Unlike traditional augmentation methods, we choose the parameters for each degradation adversarially so as to maximize system performance. By formulating image degradations in a way that is differentiable with respect to degradation parameters, we enable the use of efficient optimization methods (PGD) for choosing worst-case augmentation parameters. We demonstrate the efficacy of this method on the learning to steer task for AVs. By adversarially training against image corruptions, we produce networks that are highly robust to image corruptions. We show that the proposed differentiable augmentation schemes result in higher levels of robustness and accuracy for a range of settings as compared to baseline and state-of-the-art augmentation methods.

Trajectory planning is a key piece in the algorithmic architecture of a robot. Trajectory planners typically use iterative optimization schemes for generating smooth trajectories that avoid collisions and are optimal for tracking given the robot’s physical specifications. Starting from an initial estimate, the planners iteratively refine the solution so as to satisfy the desired constraints. In this paper, we show that such iterative optimization based planners can be vulnerable to adversarial attacks that force the planner either to fail completely, or significantly increase the time required to find a solution. The key insight here is that an adversary in the environment can directly affect the optimization cost function of a planner. We demonstrate how the adversary can adjust its own state configurations to result in poorly conditioned eigenstructure of the objective leading to failures. We apply our method against two state of the art trajectory planners and demonstrate that an adversary can consistently exploit certain weaknesses of an iterative optimization scheme.

Robots can perform various types of automated movements in the workspace. In recent years, robot applications have been expanded to a much wider scope, including robot machining, robot assembly, robot 3D printing, robot inspection, etc. Many of these applications require robots to have higher absolute accuracy compared with conventional robot part handling and welding. The capability to assess a robot’s accuracy, and further, improve accuracy becomes important. In this paper, an advanced sensor and an accompanying methodology are developed that enable manufacturers to perform accuracy assessment to improve their robot systems. A smart target (patent pending) is developed at the National Institute of Standards and Technology (NIST). This sensor is integrated with a vision-based measurement instrument to perform high accuracy measurements of six-dimensional (6-D) information (x, y, and z position, roll, pitch, and yaw orientation) for a moving robot arm. The smart target is motorized. It can constantly rotate toward the vision-based measurement instrument to maximize its line-of-sight. A use case is presented to demonstrate the accuracy degradation assessment by using the smart target on a Universal Robot.

This paper presents a manipulator-equipped unmanned aerial vehicle (UAV) performing contact-based surface inspection. Using stereo camera information, a contact point is determined and an approach path is generated to ensure an autonomous workflow. The proposed inspection method uses an impedance controller to regulate the force applied by the end-effector to the contact surface. Due to the often adverse conditions in the real world, the aerial manipulator may fail to achieve contact after approaching the surface. Therefore, the impedance controller is coupled with an online stiffness estimation algorithm to increase the overall robustness of the inspection system and achieve stable surface contact. Finally, simulations in the Gazebo environment and experiments in the real world are performed to show the feasibility of this approach.

In this paper, we study integrated estimation and control of soft robots. A significant challenge in deploying closed loop controllers is reliable proprioception via integrated sensing in soft robots. Despite the considerable advances accomplished in fabrication, modelling, and model-based control of soft robots, integrated sensing and estimation is still in its infancy. To that end, this paper introduces a new method of estimating the degree of curvature of a soft robot using a stretchable sensing skin. The skin is a spray-coated piezoresistive sensing layer on a latex membrane. The mapping from the strain signal to the degree of curvature is estimated by using a recurrent neural network. We investigate uni-directional bending as well as bi-directional bending of a single-segment soft robot. Moreover, an adaptive controller is developed to track the degree of curvature of the soft robot in the presence of dynamic uncertainties. Subsequently, using the integrated soft sensing skin, we experimentally demonstrate successful curvature tracking control of the soft robot.

We investigate improving Monte Carlo Tree Search based solvers for Partially Observable Markov Decision Processes (POMDPs), when applied to adaptive sampling problems. We propose improvements in rollout allocation, the action exploration algorithm, and plan commitment. The first allocates a different number of rollouts depending on how many actions the agent has taken in an episode. We find that rollouts are more valuable after some initial information is gained about the environment. Thus, a linear increase in the number of rollouts, i.e. allocating a fixed number at each step, is not appropriate for adaptive sampling tasks. The second alters which actions the agent chooses to explore when building the planning tree. We find that by using knowledge of the number of rollouts allocated, the agent can more effectively choose actions to explore. The third improvement is in determining how many actions the agent should take from one plan. Typically, an agent will plan to take the first action from the planning tree and then call the planner again from the new state. Using statistical techniques, we show that it is possible to greatly reduce the number of rollouts by increasing the number of actions taken from a single planning tree without affecting the agent’s final reward. Finally, we demonstrate experimentally, on simulated and real aquatic data from an underwater robot, that these improvements can be combined, leading to better adaptive sampling. The code for this work is available at https://github.com/uscresl/AdaptiveSamplingPOMCP.

Autonomous surface vessels (ASVs) always carry payloads such as passengers and cargoes. The change in the payload can sometimes be several times the weight of the vessel. The payload can cause significant changes in the dynamics of the vessel, thereby degrading the performance of the controller. This paper proposes an adaptive nonlinear model predictive control (A-NMPC) strategy for ASV trajectory tracking, which allows real-time changes in dynamics caused by severe payload variation. First, a nonlinear dynamic model that updates with the vessel’s payload is established. Then a pressure sensing method is proposed to estimate the payload of the vessel. Further, a parametric cost function that considers changing dynamics, as well as input and state constraints, is formulated in the NMPC algorithm. The tracking ability of A-NMPC is systematically studied on three different sizes of vessels in the simulation where the payload of these vessels changes eight times their inherent weight. Numerical results show that when the payload changes greatly the vessels with A-NMPC can accurately track the reference trajectory while the vessels with conventional NMPC cannot. Finally, the tracking experiments with a quarter-scale vessel in a swimming pool further verify the effectiveness of the proposed A-NMPC strategy.

Uncovering potential failure cases is a crucial step in the validation of safety critical systems such as autonomous vehicles. Failure search may be done through logging substantial vehicle miles in either simulation or real world testing. Due to the sparsity of failure events, naive random search approaches require significant amounts of vehicle operation hours to find potential system weaknesses. As a result, adaptive searching techniques have been proposed to efficiently explore and uncover failure trajectories of an autonomous policy in simulation. Adaptive Stress Testing (AST) is one such method that poses the problem of failure search as a Markov decision process and uses reinforcement learning techniques to find high probability failures. However, this formulation requires a probability model for the actions of all agents in the environment. In systems where the environment actions are discrete and dependencies among agents exist, it may be infeasible to fully characterize the distribution or find a suitable proxy. This work proposes the use of a data driven approach to learn a suitable classifier that tries to model how humans identify critical states and use this to guide failure search in AST. We show that the incorporation of critical states into the AST framework generates failure scenarios with increased safety violations in an autonomous driving policy with a discrete action space.

We consider the problem of multi-robot sensor coverage, which deals with deploying a multi-robot team in an environment and optimizing the sensing quality of the overall environment. As real-world environments involve a variety of sensory information, and individual robots are limited in their available number of sensors, successful multi-robot sensor coverage requires the deployment of robots in such a way that each individual team member’s sensing quality is maximized. Additionally, because individual robots have varying complements of sensors and both robots and sensors can fail, robots must be able to adapt and adjust how they value each sensing capability in order to obtain the most complete view of the environment, even through changes in team composition. We introduce a novel formulation for sensor coverage by multi-robot teams with heterogeneous sensing capabilities that maximizes each robot's sensing quality, balancing the varying sensing capabilities of individual robots based on the overall team composition. We propose a solution based on regularized optimization that uses sparsity-inducing terms to ensure a robot team focuses on all possible event types, and which we show is proven to converge to the optimal solution. Through extensive simulation, we show that our approach is able to effectively deploy a multi-robot team to maximize the sensing quality of an environment, responding to failures in the multi-robot team more robustly than non-adaptive approaches.

This paper aims to improve robots’ versatility and adaptability by allowing them to use a large variety of end- effector tools and quickly adapt to new tools. We propose AdaGrasp, a method to learn a single grasping policy that generalizes to novel grippers. By training on a large collection of grippers, our algorithm is able to acquire generalizable knowledge of how different grippers should be used in various tasks. Given a visual observation of the scene and the gripper, AdaGrasp infers the possible grasp poses and their grasp scores by computing the cross convolution between the shape encodings of the gripper and scene. Intuitively, this cross convolution operation can be considered as an efficient way of exhaustively matching the scene geometry with gripper geometry under different grasp poses (i.e., translations and orientations), where a good "match" of 3D geometry will lead to a successful grasp. We validate our methods in both simulation and real- world environments. Our experiment shows that AdaGrasp significantly outperforms the existing multi-gripper grasping policy method, especially when handling cluttered environments and partial observations. Code and Data are available at https://adagrasp.cs.columbia.edu.

Supervisory control of a humanoid robot in a manipulation task requires coordination of remote perception with robot action, which becomes more demanding with multiple moving cameras available for task supervision. We explore the use of autonomous camera control and selection to reduce operator workload and improve task performance in a supervisory control task. We design a novel approach to autonomous camera selection and control, and evaluate the approach in a user study which revealed that autonomous camera control does improve task performance and operator experience, but autonomous camera selection requires further investigation to benefit the operator’s confidence and maintain trust in the robot autonomy.