Jesse Levinson

Towards fully autonomous driving: Systems and algorithms

In order to achieve autonomous operation of a vehicle in urban situations with unpredictable traffic, several realtime systems must interoperate, including environment perception, localization, planning, and control. In addition, a robust vehicle platform with appropriate sensors, computational hardware, networking, and software infrastructure is essential.

Robust vehicle localization in urban environments using probabilistic maps

Autonomous vehicle navigation in dynamic urban environments requires localization accuracy exceeding that available from GPS-based inertial guidance systems. We have shown previously that GPS, IMU, and LIDAR data can be used to generate a high-resolution infrared remittance ground map that can be subsequently used for localization [4]. We now propose an extension to this approach that yields substantial improvements over previous work in vehicle localization, including higher precision, the ability to learn and improve maps over time, and increased robustness to environment changes and dynamic obstacles. Specifically, we model the environment, instead of as a spatial grid of fixed infrared remittance values, as a probabilistic grid whereby every cell is represented as its own gaussian distribution over remittance values. Subsequently, Bayesian inference is able to preferentially weight parts of the map most likely to be stationary and of consistent angular reflectivity, thereby reducing uncertainty and catastrophic errors. Furthermore, by using offline SLAM to align multiple passes of the same environment, possibly separated in time by days or even months, it is possible to build an increasingly robust understanding of the world that can be then exploited for localization. We validate the effectiveness of our approach by using these algorithms to localize our vehicle against probabilistic maps in various dynamic environments, achieving RMS accuracy in the 10cm-range and thus outperforming previous work. Importantly, this approach has enabled us to autonomously drive our vehicle for hundreds of miles in dense traffic on narrow urban roads which were formerly unnavigable with previous localization methods.

Map-Based Precision Vehicle Localization in Urban Environments

Many urban navigation applications (e.g., autonomous navigation, driver assistance systems) can benefit greatly from localization with centimeter accuracy. Yet such accuracy cannot be achieved reliably with GPS-based inertial guidance systems, specifically in urban settings. We propose a technique for high-accuracy localization of moving vehicles that utilizes maps of urban environments. Our approach integrates GPS, IMU, wheel odometry, and LIDAR data acquired by an instrumented vehicle, to generate high-resolution environment maps. Offline relaxation techniques similar to recent SLAM methods [2, 10, 13, 14, 21, 30] are employed to bring the map into alignment at intersections and other regions of self-overlap. By reducing the final map to the flat road surface, imprints of other vehicles are removed. The result is a 2-D surface image of ground reflectivity in the infrared spectrum with 5cm pixel resolution. To localize a moving vehicle relative to these maps, we present a particle filter method for correlating LIDAR measurements with this map. As we show by experimentation, the resulting relative accuracies exceed that of conventional GPS-IMU-odometry-based methods by more than an order of magnitude. Specifically, we show that our algorithm is effective in urban environments, achieving reliable real-time localization with accuracy in the 10centimeter range. Experimental results are provided for localization in GPS-denied environments, during bad weather, and in dense traffic. The proposed approach has been used successfully for steering a car through narrow, dynamic urban roads.

Towards 3D object recognition via classification of arbitrary object tracks

Object recognition is a critical next step for autonomous robots, but a solution to the problem has remained elusive. Prior 3D-sensor-based work largely classifies individual point cloud segments or uses class-specific trackers. In this paper, we take the approach of classifying the tracks of all visible objects. Our new track classification method, based on a mathematically principled method of combining log odds estimators, is fast enough for real time use, is non-specific to object class, and performs well (98.5% accuracy) on the task of classifying correctly-tracked, well-segmented objects into car, pedestrian, bicyclist, and background classes. We evaluate the classifiers performance using the Stanford Track Collection, a new dataset of about 1.3 million labeled point clouds in about 14,000 tracks recorded from an autonomous vehicle research platform. This dataset, which we make publicly available, contains tracks extracted from about one hour of 360-degree, 10Hz depth information recorded both while driving on busy campus streets and parked at busy intersections.

Unsupervised Calibration for Multi-beam Lasers

Light Detection and Ranging (LIDAR) sensors have become increasingly common in both industrial and robotic applications. LIDAR sensors are particularly desirable for their direct distance measurements and high accuracy, but traditionally have been configured with only a single rotating beam. However, recent technological progress has spawned a new generation of LIDAR sensors equipped with many simultaneous rotating beams at varying angles, providing at least an order of magnitude more data than single-beam LIDARs and enabling new applications in mapping [6], object detection and recognition [15], scene understanding [16], and SLAM [9].

Traffic light mapping, localization, and state detection for autonomous vehicles

Detection of traffic light state is essential for autonomous driving in cities. Currently, the only reliable systems for determining traffic light state information are non-passive proofs of concept, requiring explicit communication between a traffic signal and vehicle. Here, we present a passive camera-based pipeline for traffic light state detection, using (imperfect) vehicle localization and assuming prior knowledge of traffic light location. First, we introduce a convenient technique for mapping traffic light locations from recorded video data using tracking, back-projection, and triangulation. In order to achieve robust real-time detection results in a variety of lighting conditions, we combine several probabilistic stages that explicitly account for the corresponding sources of sensor and data uncertainty. In addition, our approach is the first to account for multiple lights per intersection, which yields superior results by probabilistically combining evidence from all available lights. To evaluate the performance of our method, we present several results across a variety of lighting conditions in a real-world environment. The techniques described here have for the first time enabled our autonomous research vehicle to successfully navigate through traffic-light-controlled intersections in real traffic.

Automatic Online Calibration of Cameras and Lasers

The combined use of 3D scanning lasers with 2D cameras has become increasingly popular in mobile robotics, as the sparse depth measurements of the former augment the dense color information of the latter. Sensor fusion requires precise 6DOF transforms between the sensors, but hand-measuring these values is tedious and inaccurate. In addition, autonomous robots can be rendered inoperable if their sensors’ calibrations change over time. Yet previously published camera-laser calibration algorithms are offline only, requiring significant amounts of data and/or specific calibration targets; they are thus unable to correct calibration errors that occur during live operation. In this paper, we introduce two new real-time techniques that enable camera-laser calibration online, automatically, and in arbitrary environments. The first is a probabilistic monitoring algorithm that can detect a sudden miscalibration in a fraction of a second. The second is a continuous calibration optimizer that adjusts transform offsets in real time, tracking gradual sensor drift as it occurs. Although the calibration objective function is not globally convex and cannot be optimized in real time, in practice it is always locally convex around the global optimum, and almost everywhere else. Thus, the local shape of the objective function at the current parameters can be used to determine whether the sensors are calibrated, and allows the parameters to be adjusted gradually so as to maintain the global optimum. In several online experiments on thousands of frames in real markerless scenes, our method automatically detects miscalibrations within one second of the error exceeding .25 deg or 10cm, with an accuracy of 100%. In addition, rotational sensor drift can be tracked in real-time with a mean error of just .10 deg. Together, these techniques allow significantly greater flexibility and adaptability of robots in unknown and potentially harsh environments.

Precision tracking with sparse 3D and dense color 2D data

Precision tracking is important for predicting the behavior of other cars in autonomous driving. We present a novel method to combine laser and camera data to achieve accurate velocity estimates of moving vehicles. We combine sparse laser points with a high-resolution camera image to obtain a dense colored point cloud. We use a color-augmented search algorithm to align the dense color point clouds from successive time frames for a moving vehicle, thereby obtaining a precise estimate of the tracked vehicles velocity. Using this alignment method, we obtain velocity estimates at a much higher accuracy than previous methods. Through pre-filtering, we are able to achieve near real time results. We also present an online method for real-time use with accuracies close to that of the full method. We present a novel approach to quantitatively evaluate our velocity estimates by tracking a parked car in a local reference frame in which it appears to be moving relative to the ego vehicle. We use this evaluation method to automatically quantitatively evaluate our tracking performance on 466 separate tracked vehicles. Our method obtains a mean absolute velocity error of 0.27 m/s and an RMS error of 0.47 m/s on this test set. We can also qualitatively evaluate our method by building color 3D car models from moving vehicles. We have thus demonstrated that our method can be used for precision car tracking with applications to autonomous driving and behavior modeling.

Combining 3D Shape, Color, and Motion for Robust Anytime Tracking

Although object tracking has been studied for decades, real-time tracking algorithms often suffer from low accuracy and poor robustness when confronted with difficult, realworld data. We present a tracker that combines 3D shape, color (when available), and motion cues to accurately track moving objects in real-time. Our tracker allocates computational effort based on the shape of the posterior distribution. Starting with a coarse approximation to the posterior, the tracker successively refines this distribution, increasing in tracking accuracy over time. The tracker can thus be run for any amount of time, after which the current approximation to the posterior is returned. Even at a minimum runtime of 0.7 milliseconds, our method outperforms all of the baseline methods of similar speed by at least 10%. If our tracker is allowed to run for longer, the accuracy continues to improve, and it continues to outperform all baseline methods. Our tracker is thus anytime, allowing the speed or accuracy to be optimized based on the needs of the application.

A probabilistic framework for car detection in images using context and scale

Detecting cars in real-world images is an important task for autonomous driving, yet it remains unsolved. The system described in this paper takes advantage of context and scale to build a monocular single-frame image-based car detector that significantly outperforms the baseline. The system uses a probabilistic model to combine multiple forms of evidence for both context and scale to locate cars in a real-world image. We also use scale filtering to speed up our algorithm by a factor of 3.3 compared to the baseline. By using a calibrated camera and localization on a road map, we are able to obtain context and scale information from a single image without the use of a 3D laser. The system outperforms the baseline by an absolute 9.4% in overall average precision and 11.7% in average precision for cars smaller than 50 pixels in height, for which context and scale cues are especially important.