A Novel Learning Framework for Sampling-Based Motion Planning in Autonomous Driving

To compute collision-free and dynamically-feasibile trajectories that satisfy high-level specifications given in a planning-domain definition language, this paper proposes to combine sampling-based motion planning with symbolic action planning. The proposed approach, Sampling-based Motion and Symbolic Action Planner (SMAP), leverages from sampling-based motion planning the underlying idea of searching for a solution trajectory by selectively sampling and exploring the continuous space of collision-free and dynamically-feasible motions. Drawing from AI, SMAP uses symbolic action planning to identify actions and regions of the continuous space that sampling-based motion planning can further explore to significantly advance the search. The planning layers interact with each-other through estimates on the utility of each action, which are computed based on information gathered during the search. Simulation experiments with dynamical models of vehicles carrying out tasks given by high-level STRIPS specifications provide promising initial validation, showing that SMAP efficiently solves challenging problems.

We discuss the optimality and computational efficiency of sampling based motion planning (SBMP), which calculates dynamically precise and approximately optimal state transitions using arbitrarily selected nonlinear control system models. To ensure high optimality and computational efficiency, SBMP requires an approximately uniform state sampling function, though the non-linearity of the system models does not allow a perfect function. We propose linear prediction-based uniform state sampling (LPUSS) that samples approximately uniform state points while ensuring a dynamically correct state transition profile with a small calculation cost. LPUSS samples a state by using the given non-linear control system model after determining the input values by using a local linear transition model. We developed a mechanical motion planning system using LPUSS, articulated body algorithm, and parallel computing techniques. To validate LPUSS, we conducted experiments on double, triple, and sixtuple inverted pendulum models. LPUSS showed better optimality and computational efficiency with the double and triple inverted pendulum models, compared with randomized kinodynamic planning (RKP), which is based on rapid random tree (RRT), and our previously proposed rapid semi-optimal motion-planning method in which state sampling is based on uniform inputs. In particular, compared with our previous method, LPUSS was respectively 130 times and 3,000 times faster on double and triple inverted pendulum models under the condition of the same optimality. LPUSS found an approximately optimal swing up motion for the sixtuple inverted pendulum model within 40 minutes. According to our survey, there is no other optimization method that is applicable to higher than quadruple inverted pendulum models with standard constraints.

In this paper, we introduce a novel and efficient hybrid trajectory planning method for autonomous driving in highly constrained environments. The contributions of this paper are fourfold. First, we present a trajectory planning framework that is able to handle geometry constraints, nonholonomic constraints, and dynamics constraints of cars in a humanlike and layered fashion and generate curvature-continuous, kinodynamically feasible, smooth, and collision-free trajectories in real time. Second, we present a derivative-free global path modification algorithm to extract high-order state information in free space for state sampling. Third, we extend the regular state-space sampling method widely used in on-road autonomous driving systems to a multi-phase deterministic state-space sampling method that is able to approximate complex maneuvers. Fourth, we improve collision checking accuracy and efficiency by using a different car footprint approximation strategy and a two-phase collision checking routine. A range of challenging simulation experiments show that the proposed method returns high-quality trajectories in real time and outperforms existing planners, such as hybrid A* and conjugate-gradient descent path smoother in terms of path quality, efficiency, and computation resources used.

We present a robust teach-and-repeat system enabling an autonomous car to carry out transportation tasks in non-urban environments. Our vehicle first learns a desired route between two arbitrary places by following a human guide tracked by a marker-less, LiDAR-based object tracking algorithm. By moving backward along the route, the guide has the ability to make the vehicle correct the recorded path while it is still navigating autonomously. In the second phase, the car starts shuttling between the route's end points. To this end, a Hybrid A∗-based planner generates dynamically feasible, collision-free trajectories whose costs are based on a multitude of features of the local environment, such as slopes, vegetation and road probabilities. A hierarchical state machine coordinates the interaction between the various software modules and monitors the mission's progress. When it detects a road blockage during shuttling, it uses Open Street Map data to find an alternative route to the current goal. We tested the system extensively on our autonomous car MuCAR-3. Apart from data recorded during test runs, we also present the results of our participation in the European Land Robot Trial (ELROB) 2016 competition, where MuCAR-3 took first place in the MULE scenario.

Autonomous cars operate following specific commands, routines and sequences based on algorithms developed by humans. When they will appear on the market, they shall coexist with traditional cars driven by human drivers. What data shall be available and what data shall they collect about these cars and drivers, and how shall they process them? What are the boundaries of what information can be recorded for later use, and is it possible and/or necessary to regulate the Artificial Intelligence programs that control and command the operations of autonomous cars to prevent misuse of personal data, so that the privacy of drivers shall be protected?.

The emergence of autonomous driving is becoming a reality worldwide. However, in emergency situations, human drivers must assume control of the vehicle, hence detecting driver vigilance is critical. Although many electroencephalographic studies have documented alterations in brain rhythms associated with fatigue, the results have been inconsistent. Furthermore, in high-level automated driving contexts, the driver does not operate the vehicle except in emergencies. In this study, magnetoencephalograms from eight healthy participants were recorded as they watched a 20-minute video from the driver’s perspective. The participants were instructed to press a button when a checkered flag appeared, simulating their response to an emergency. Statistical analysis demonstrated that response time to the checkered flag was prolonged during the End period (final third) of the experiment (p = 0.04). The increase in reaction time from the Beginning (first third) to the Middle (middle third) period correlated negatively with alpha-band activities in the occipital cortices during all three periods (Beginning, r = −0.72; p = 0.04; Middle, r = −0.73, p = 0.04; End r = −0.78, p = 0.02). In other words, participants with higher alpha-band activity in the Beginning period did not experience prolonged reaction time toward the Middle period. This result suggests that alpha-band activity in the occipital cortex may serve as an indicator of whether participants maintain vigilance during autonomous driving.

Investigating adaptive sampling strategies for optimal building energy performance using artificial neural networks and kriging surrogate models

Maritime transportation is vital to the global economy. With the increased operating and labor costs of maritime transportation, autonomous shipping has attracted much attention in both industry and academia. Autonomous shipping can not only reduce the marine accidents caused by human factors but also save labor costs. Path planning is one of the key technologies to enable the autonomy of ships. However, mainstream ship path planning focuses on searching for the shortest path and controlling the vehicle in order to track it. Such path planning methods may lead to a dynamically infeasible trajectory that fails to avoid obstacles or reduces fuel efficiency. This paper presents a data-driven, efficient, and safe path planning (ESP) method that considers ship dynamics to provide a real-time optimal trajectory generation. The optimization objectives include fuel consumption and trajectory smoothness. Furthermore, ESP is capable of fast replanning when encountering obstacles. ESP consists of three components: (1) A path search method that finds an optimal search path with the minimum number of sharp turns from the geographic data collected by the geographic information system (GIS); (2) a minimum-snap trajectory optimization formulation with dynamic ship constraints to provide a smooth and collision-free trajectory with minimal fuel consumption; (3) a local trajectory replanner based on B-spline to avoid unexpected obstacles in real time. We evaluate the performance of ESP by data-driven simulations. The geographical data have been collected and updated from GIS. The results show that ESP can plan a global trajectory with safety, minimal turning points, and minimal fuel consumption based on the maritime information provided by nautical charts. With the long-range perception of onboard radars, the ship can avoid unexpected obstacles in real time on the planned global course.

To better understand the road condition and make correct driving decisions, traffic sign recognition becomes a crucial component commonly equipped in the vision system of modern autonomous cars. The state-of-the-art traffic sign recognition models are designed with the backbones of deep neural networks (DNNs) since DNNs are powerful to extract more effective visual features that benefit recognition performance. As the recent studies on adversarial attacks have shown that DNNs are easy to be fooled by perturbed images and lead to misclassification, in this article, we explore the vulnerability of the DNN-based traffic sign recognition model. Most existing adversarial attack methods limitedly focus on the white-box attack on the recognition models whose underlying configurations (e.g., network architectures and parameters) are accessible. Differently, we propose a novel attacking method dubbed adaptive square attack (ASA) that can accomplish the black-box attack, i.e., bypassing the access the configurations of the recognition models. Specifically, the proposed ASA method employs an efficient sampling strategy that can generate perturbations for traffic sign images with fewer query times. Extensive experiments on the benchmark data set German traffic sign recognition benchmark with large-scale traffic sign images for autonomous cars show that our proposed ASA method is advanced to perform the black-box attack with high efficiency. Although the generated adversarial traffic sign images by the proposed ASA method are visually similar to the raw images with almost imperceptible differences, they can successfully lead to the misclassification of the state-of-the-art recognition model.

Juvenile dermatomyositis (JDM) is a rare systemic autoimmune disease specifically affecting children. Mycophenolate mofetil (MMF) is an immunosuppressant used to treat JDM. Mycophenolic acid (MPA) is an active metabolite of MMF. This study aimed to develop a population pharmacokinetic (PPK) model of MPA in children with JDM and optimize the limited sampling strategy (LSS). Fifteen JDM patients treated with MMF, at a median age of 7.35 (range, 3.09–16.1) years, were included. Blood samples were collected at 30 minutes pre-dose, 20 minutes, 60 minutes and 180 minutes post-dose to measure the MPA concentrations. Data were retrospectively collected from the electronic medical records. A two-compartment model with first-order absorption, lag time in absorption, and first-order elimination was developed. Height and co-administered cotrimoxazole were added as the covariates to the model. Concentrations at different time points were simulated and area under the concentration–time curve (AUC0–12 h) was calculated. By removing one sampling point at a time, AUC0–12 h from three-point sampling strategy was re-calculated via Bayesian approach. AUC0–12 h from the three-point sampling strategy (by removing the point at 20 minutes post-dose) had the strongest correlation with AUC0–12 h from the four-point sampling strategy (Pearson’s r = 0.971).

Advances in information and signal processing technologies have a significant impact on autonomous driving (AD), improving driving safety while minimizing the efforts of human drivers with the help of advanced artificial intelligence (AI) techniques. Recently, deep learning (DL) approaches have solved several real-world problems of complex nature. However, their strengths in terms of control processes for AD have not been deeply investigated and highlighted yet. This survey highlights the power of DL architectures in terms of reliability and efficient real-time performance and overviews state-of-the-art strategies for safe AD, with their major achievements and limitations. Furthermore, it covers major embodiments of DL along the AD pipeline including measurement, analysis, and execution, with a focus on road, lane, vehicle, pedestrian, drowsiness detection, collision avoidance, and traffic sign detection through sensing and vision-based DL methods. In addition, we discuss on the performance of several reviewed methods by using different evaluation metrics, with critics on their pros and cons. Finally, this survey highlights the current issues of safe DL-based AD with a prospect of recommendations for future research, rounding up a reference material for newcomers and researchers willing to join this vibrant area of Intelligent Transportation Systems.

With the progress of autonomous driving technology, obtaining a safe and smooth trajectory in complex environments is the focus of motion planning in recent years. This paper presents a post-optimization method based on the gradient descent and Bezier curve, which can obtain a safer and more comfortable trajectory in the static obstacle scene. The optimized path is farther away from obstacles locally and can be tracked by the low-controller. Our post-optimization method has two main steps which consider the factors of collision avoidance and smoothing separately. The first step uses the gradient descent method to optimize so that the path is farther away from obstacles. The cost function of the gradient descent method mainly considers obstacle punishment. The second step uses the variable order Bezier curve to smooth segments which have sharp inflection points caused by the gradient descent. The simulation experiments have been performed to evaluate the feasibility and efficiency of the proposed method. The simulation results show that our algorithm can achieve the purpose of staying away from obstacles locally and refine the vehicle smoothness as well.

Autonomous driving is currently the focus of considerable media attention, and the automotive industry is working on harnessing technological advances to accelerate development. In doing so, different levels of automated driving must be mastered (Table 1) and defined [1]. At higher automation levels, the execution, monitoring, and fallback performance are increasingly handled by the system and not the human driver, resulting in more demanding requirements for the sensor system. As shown in [2], radar sensors are a key technology to enable Society of Automotive Engineers level 5, full automation.

Planning autonomous driving behaviors in dense traffic is challenging. Human drivers are able to influence their road environment to achieve (otherwise unachievable) goals, by communicating their intents to other drivers. An autonomous system that is required to drive in the presence of human traffic must thus possess this fundamental negotiation capability. This work presents a novel benchmark that includes a stochastic driver negotiation model and a framework for training policies to drive and negotiate based on reinforcement learning. It is shown that driving policies trained in this framework lead to greater safety, higher mission accomplishment rates and more driving comfort, and can generalize across scenarios.

Since human drivers only consider the driving-related factors that affect vehicle control depending on the situation, they can drive safely even in diverse driving environments. To mimic this behavior, we propose an autonomous driving framework based on the two-stage representation learning that initially splits the latent features as domain-specific features and domain-general features. Subsequently, the dynamic-object features, which contain information of dynamic objects, are disentangled from latent features using mutual information estimator. In this study, the problem in behavior cloning is divided into several domain-specific subspaces, with experts becoming specialized on each domain-specific policy. The proposed mixture of domain-specific experts (MoDE) model predicts the final control values through the cooperation of experts using a gating function. The domain-specific features are used to calculate the importance weight of the domain-specific experts, and the disentangled domain-general and dynamic-object features are applied in estimating the control values. To validate the proposed MoDE model, we conducted several experiments and achieved a higher success rate on the CARLA benchmarks under several conditions and tasks than state-of-the-art approaches.