IROS 2019 Papers

We envision a system that concisely describes the rules of air traffic control, assists human operators and supports dense autonomous air traffic around commercial airports. We develop a method to learn the rules of air traffic control from real data as a cost function via maximum entropy inverse reinforcement learning. This cost function is used as a penalty for a search-based motion planning method that discretizes both the control and the state space. We illustrate the methodology by showing that our approach can learn to imitate the airport arrival routes and separation rules of dense commercial air traffic. The resulting trajectories are shown to be safe, feasible, and efficient.

The method of Frobenius Norm (FN) minimization of forward kinematic Jacobians is presented to minimize the sensitivity of a robotic manipulator. We demonstrate the effectiveness of this approach with a Monte Carlo simulation of an Assembler robot. The Assembler is described as a stack of Stewart Platforms designed for in-space assembly to aide NASA in deep space exploration. The translations and rotations between each Stewart Platform define the forward kinematics of the Assember, which analytically determine the end effector position and orientation. However, selecting the poses of each Stewart Platform which yield a desired end effector state, defined as the inverse kinematics, is an underconstrained nonlinear optimization problem for large Assembler stacks.

Inverse dynamics modeling is a critical problem for the computed-torque control of robotic manipulator. This paper presents a novel recurrent network based on the modified Simple Recurrent Unit (SRU) with hierarchical memory (SRU-HM), which is achieved by the nested SRU structure. In this way, it enables the capability to retain the long-term information in the distant past, compared with the conventional stacked structure. The hidden state of SRU is able to provide more complete information relevant to current prediction. Experimental results demonstrate that the proposed method can improve the accuracy of dynamics model greatly, and outperforms the state-of-the-art methods.

Robotics for hazardous environments is currently an important area of research, with the ambition of reducing human risk in potentially devastating situations. Here, we are presenting a Modular Control Framework (MCF) for a low-cost robot with limited sensory resources to address this issue. As a proof of concept, we emulate 3 scenarios - (1) adaptive planning for obstruction avoidance (road block), (2) object identification and support-case-based behaviour adjustment (search and rescue) and (3) autonomous navigation through the environment with reporting of structural status (patrol and monitoring). These were implemented and validated using a Cozmo robot in a small-scale Lego environment. We found that our system can reroute in 90%, can help an injured person 80% and report about failing equipment in 80% of all tested cases, where most of the fails were caused by the object detection used. Our MCF is implemented using ROS, making it easy to use and adjust for other robotic platforms.

Flexible manufacturing processes demand robots to easily adapt to changes in the environment and interact with humans. In such dynamic scenarios, robotic tasks may be programmed through learning-from-demonstration (LfD) approaches, where a nominal plan of the task is learned by the robot. However, the learned plan may need to be adapted in order to fulfill additional requirements or overcome unexpected environment changes. When the required adaptation occurs at the end-effector trajectory level, a human operator may want to intuitively show the robot the desired changes by physically interacting with it. In this scenario, the robot needs to understand the human intended changes from noisy haptic data, quickly adapt accordingly and execute the nominal task plan when no further adaptation is needed. This paper addresses the aforementioned challenges by leveraging LfD and Bayesian optimization to endow the robot with data-efficient adaptation capabilities. Our approach exploits the sensed interaction forces to guide the robot adaptation, and speeds up the optimization process by defining local search spaces extracted from the learned task model. We show how our framework quickly adapts the learned spatial-temporal patterns of the task, leading to deformed trajectory distributions that are consistent with the nominal plan and the changes introduced by the human.

In order to drive safely and efficiently under merging scenarios, autonomous vehicles should be aware of their surroundings and make decisions by interacting with other road participants. Moreover, different strategies should be made when the autonomous vehicle is interacting with drivers having different level of cooperativeness. Whether the vehicle is on the merge-lane or main-lane will also influence the driving maneuvers since drivers will behave differently when they have the right-of-way than otherwise. Many traditional methods have been proposed to solve decision making problems under merging scenarios. However, these works either are incapable of modeling complicated interactions or require implementing hand-designed rules which cannot properly handle the uncertainties in real-world scenarios. In this paper, we proposed an interaction-aware decision making with adaptive strategies (IDAS) approach that can let the autonomous vehicle negotiate the road with other drivers by leveraging their cooperativeness under merging scenarios. A single policy is learned under the multi-agent reinforcement learning (MARL) setting via the curriculum learning strategy, which enables the agent to automatically infer other drivers' various behaviors and make decisions strategically. A masking mechanism is also proposed to prevent the agent from exploring states that violate common sense of human judgment and increase the learning efficiency. An exemplar merging scenario was used to implement and examine the proposed method.

Integer programming (IP) has proven to be highly effective in solving many path-based optimization problems in robotics. However, the applications of IP are generally done in an ad-hoc, problem-specific manner. In this work, after examined a wide range of path-based optimization problems, we describe an IP solution methodology for these problems that is both easy to apply (in two simple steps) and high-performance in terms of the computation time and the achieved optimality. We demonstrate the generality of our approach through the application to three challenging path-based optimization problems: multi-robot path planning(MPP), minimum constraint removal(MCR), and reward collection problems(RCPs). Associated experiments show that the approach can efficiently produce (near-)optimal solutions for problems with large state spaces, complex constraints, and complicated objective functions. In conjunction with the proposition of the IP methodology, we introduce two new and practical robotics problems: multi-robot minimum constraint removal(MMCR) and multi-robot path planning(MPP) with partial solutions, which can be quickly and effectively solved using our proposed IP solution pipeline.

Current perception systems mostly require direct line of sight to anticipate and ultimately prevent potential collisions at intersections with other road users. We present a fully integrated autonomous system capable of detecting shadows or weak illumination changes on the ground caused by a dynamic obstacle in NLoS scenarios. This additional virtual sensor “ShadowCam” extends the signal range utilized so far by computer-vision ADASs. We show that (1) our algorithm maintains the mean classification accuracy of around 70% even when it doesn't rely on infrastructure - such as AprilTags - as an image registration method. We validate (2) in real-world experiments that our autonomous car driving in night time conditions detects a hidden approaching car earlier with our virtual sensor than with the front facing 2-D LiDAR.

Deep Learning-based object detectors enhance the capabilities of remote sensing platforms, such as Unmanned Aerial Vehicles (UAVs), in a wide spectrum of machine vision applications. However, the integration of deep learning introduces heavy computational requirements, preventing the deployment of such algorithms in scenarios that impose low-latency constraints during inference, in order to make mission-critical decisions in real-time. In this paper, we address the challenge of efficient deployment of region-based object detectors in aerial imagery, by introducing an informed methodology for extracting candidate detection regions (proposals). Our approach considers information from the UAV on-board sensors, such as flying altitude and light-weight computer vision filters, along with prior domain knowledge to intelligently decrease the number of region proposals by eliminating false-positives at an early stage of the computation, reducing significantly the computational workload while sustaining the detection accuracy. We apply and evaluate the proposed approach on the task of vehicle detection. Our experiments demonstrate that state-of-the-art detection models can achieve up to 2.6x faster inference by employing our altitude-aware data-driven methodology. Alongside, we introduce and provide to the community a novel vehicle-annotated and altitude-stamped dataset of real UAV imagery, captured at numerous flying heights under a wide span of traffic scenarios.

Building occupancy maps of the environment is a fundamental problem for robot autonomy. A common assumption in early work was that the occupancy states of different map elements are independent. Recently, Gaussian Process (GP) techniques were proposed to capture correlation, which is important not only for improved accuracy but also for uncertainty quantification and autonomous exploration based on the predicted occupancy of nearby unexplored areas. Despite these desirable properties, current GP mapping techniques are limited to small maps and slow inference speeds. This paper proposes an information space formulation of the GP mapping problem. If a decomposable radial kernel is evaluated over a latent grid of pseudo-input points, the resulting kernel matrix has a Kronecker-product-of-Toeplitz-matrices structure that allows very efficient representation of the occupancy distribution. We utilize this structure to design an information filter occupancy mapping algorithm with linear time and memory complexity that still permits continuous space observations and predictions.

Positioning accuracy of cable-driven parallel robots is influenced by many factors such as geometry, actuator sensor accuracy and disturbances in the applied wrench. Another uncertainty source is the elasticity of the cables. While the influence of many factors may be decreased by calibration and/or sensor fusion, elasticity parameters are difficult to estimate and their effect on the positioning errors has yet to be determined. In this paper we consider a generic cable model that includes cable elasticity and the effect of cable weight and we propose a generic algorithm that allows one to safely calculate the minimum and maximum of the positioning error at a given pose when the elasticity parameters are constrained to lie within some given bounds. The algorithm is designed for being able to manage the effect of different uncertainties sources and we compare the influence of elasticity versus the effect of uncertainties in the cable lengths.

When building a geometric scene understanding system for autonomous vehicles, it is crucial to know when the system might fail. Most contemporary approaches cast the problem as depth regression, whose output is a depth value for each pixel. Such approaches cannot diagnose when failures might occur. One attractive alternative is a deep Bayesian network, which captures uncertainty in both model parameters and ambiguous sensor measurements. However, estimating uncertainties is often slow and the distributions are often limited to be uni-modal. In this paper, we recast the continuous problem of depth regression as discrete binary classification, whose output is an un-normalized distribution over possible depths for each pixel. Such output allows one to reliably and efficiently capture multi-modal depth distributions in ambiguous cases, such as depth discontinuities and reflective surfaces. Results on standard benchmarks show that our method produces accurate depth predictions and significantly better uncertainty estimations than prior art while running near real-time. Finally, by making use of uncertainties of the predicted distribution, we significantly reduce streak-like artifacts and improves accuracy as well as memory efficiency in 3D map reconstruction. Video and code can be found on the project website1.

Shared control methodologies have the potential of enabling wheelchair-bound users with limited motor abilities to perform tasks that would usually be beyond their capabilities. Deriving such methodologies in advance is challenging, since they are frequently heavily dependent on unique characteristics of users. Learning Assistance by Demonstration paradigms allow derivation of customized policies by recording how remote human assistants assist particular users. However, for accurate determination of the optimal policies for each user and context, the remote assistant needs to infer the intention of the driver, which is frequently obscured by noisy signals dependent on the user's motor impairment. In this paper, we propose a multimodal teleoperation interface, incorporating map information, haptic feedback and user eye-gaze data, and examine which of these factors are most important for allowing accurate determination of user intention in a simulated tremor experiment. Our study indicates that, for expert assistants, presence of additional haptic and gaze information increases their ability to accurately infer the user's intention, providing supporting evidence for the utility of multimodal interfaces in remote assistance scenarios for Learning Assistance by Demonstration. Our study also reveals strong individual preferences on the different modalities, with large variations of performance occurring depending on whether supplemental eye-gaze or haptic information was given.

Smart coaching platforms are emerging which combine BodySensor-Networks with AI-based training software to monitor and analyze body motions of athletes, workers, or medical patients. This allows for new opportunities to explore algorithms to interpret body sensor data and provide analytical feedback for learning a physical task, refining body motions, or to protect from work-related injuries. This paper presents a solution to non-invasively equip a person with sensors of a Smart Training System (STS) to improve training efficiency during sport activities. Our system calculates the significance of each body part during physical activities and provides targeted feedback on which body locations are under-performing. In experiments, the system collected data from 13 inertial sensors attached to the entire body of inexperienced golf learners. Using an indoor golf training net with a central target with 3 concentric zones, 1,080 real-world golf swings of 11 participants were analyzed. During the first 30 swings of each participant, the system learned distributions of motions from each sensor, conditioned on swing performance reported by users from their hitting location on the target. In the later 70 swings, feedback was provided to a subgroup of 8 participants, by computing, for an optimal set of features determined during training, the largest discrepancy. The remaining 3 (control) participants received no feedback. From only 100 golf swings for each participant, our system led to significantly improved scores by on average 3.7x (t-test, p <; 0.0001) over the latter 70 swings. Our results suggest that the combination of motion sensors and processing developed here was able to yield significantly improved golf swing training.

Animals in nature, such as geckos, inchworms, and felines can climb on various surfaces using different mechanisms and serve as references for the study of bio-inspired robots. This paper presents an inchworm-inspired climbing robot that consists of soft body and feet. The soft robot is actuated by shape memory alloy wires and utilizes microspine arrays to attach its feet to rough or soft surfaces. A series of experiments to test the functionality of the feet and torso of the designed robot have verified the theoretical feasibility of the robot. Results have shown that the designed bio-inspired robot can climb on inclined or vertical curved surfaces and flat surfaces. The robot can also adapt to the underwater environment. Thus, this robot has great potential for various applications such as pipeline inspection.

Learning from Demonstration depends on a robot learner generalising its learned model to unseen conditions. Generally, it is not feasible for a person to provide demonstrations that account for all possible variations in non-trivial tasks. While there are many learning methods that can handle interpolation of observed data effectively, extrapolation from observed data offers a much greater challenge. To address this problem, this paper proposes a modified Task-Parameterised Gaussian Mixture Regression method that considers the relevance of task parameters during trajectory generation, as determined by variance in the data. The benefits of the proposed method are first explored using a simulated reaching task data set. Here it is shown that the proposed method offers far-reaching, low-error extrapolation abilities that are different in nature to existing learning methods. Data collected from novice users for a real-world manipulation task is then considered, where it is shown that the proposed method is able to effectively reduce grasping performance errors by \sim \sim 40% and extrapolate to unseen grasp targets under real-world conditions. These results indicate the proposed method can benefit novice users by placing less reliance on the user to provide high quality demonstration data sets.

We use static object data to improve success detection for stacking objects on and nesting objects in one another. Such actions are necessary for certain robotics tasks, e.g., clearing a dining table or packing a warehouse bin. However, using an RGB-D camera to detect success can be insufficient: same-colored objects can be difficult to differentiate, and reflective silverware cause noisy depth camera perception. We show that adding static data about the objects themselves improves the performance of an end-to-end pipeline for classifying action outcomes. Images of the objects, and language expressions describing them, encode prior geometry, shape, and size information that refine classification accuracy. We collect over 13 hours of egocentric manipulation data for training a model to reason about whether a robot successfully placed unseen objects in or on one another. The model achieves up to a 57% absolute gain over the task baseline on pairs of previously unseen objects.

Local trajectory optimisation techniques are a powerful tool for motion planning. However, they often get stuck in local optima depending on the quality of the initial solution and consequently, often do not find a valid (i.e. collision free) trajectory. Moreover, they often require fine tuning of a cost function to obtain the desired motions. In this paper, we address both problems by combining local trajectory optimisation with learning from demonstrations. The human expert demonstrates how to reach different target end-effector locations in different ways. From these demonstrations, we estimate a trajectory distribution, represented by a Probabilistic Movement Primitive (ProMP). For a new target location, we sample different trajectories from the ProMP and use these trajectories as initial solutions for the local optimisation. As the ProMP generates versatile initial solutions for the optimisation, the chance of finding poor local minima is significantly reduced. Moreover, the learned trajectory distribution is used to specify the smoothness costs for the optimisation, resulting in solutions of similar shape as the demonstrations. We demonstrate the effectiveness of our approach in several complex obstacle avoidance scenarios.

Visual odometry is an essential problem for mobile robots. Traditional methods for solving VO mostly utilize geometric optimization. While capable of achieving high accuracy, these methods require accurate sensor calibration and complicated parameter tuning to work well in practice. With the rise of deep learning, there has been increased interest in the end-to-end, learning-based methods for VO, which have the potential to improve robustness. However, learning-based methods for VO so far are less accurate than geometric methods. We argue that one of the main issues is that the current ego-motion estimation task is different from other problems where deep learning has been successful such as object detection. We define a novel cost function for learning-based VO considering the mathematical properties of the group homomorphism. In addition to the standard L2 loss, we incorporate losses based on the identity, inverse and closure properties of SE(3) rigid motion. Furthermore, we propose to reduce the VO drift by estimating the drivable regions using semantic segmentation and incorporate this information into a pose graph optimization. Experiments on KITTI datasets show that the novel cost function can improve ego-motion estimation compared to the state-of-the-art and the drivable region-based correction further reduces the VO drift.