Lightning NeRF: Efficient Hybrid Scene Representation for Autonomous Driving

Recent studies have highlighted the promising application of NeRF in autonomous driving contexts. However, the complexity of outdoor environments, combined with the restricted viewpoints in driving scenarios, complicates the task of precisely reconstructing scene geometry. Such challenges often lead to diminished quality in reconstructions and extended durations for both training and rendering. To tackle these challenges, we present Lightning NeRF. It uses an efficient hybrid scene representation that effectively utilizes the geometry prior from LiDAR in autonomous driving scenarios. Lightning NeRF significantly improves the novel view synthesis performance of NeRF and reduces computational overheads. Through evaluations on real-world datasets, such as KITTI-360, Argoverse2, and our private dataset, we demonstrate that our approach not only exceeds the current state-of-the-art in novel view synthesis quality but also achieves a five-fold increase in training speed and a ten-fold improvement in rendering speed. Codes are available at https://github.com/VISION-SJTU/Lightning-NeRF .

EAST: Environment Aware Safe Tracking using Planning and Control Co-Design

This paper considers the problem of autonomous robot navigation in unknown environments with moving obstacles. We propose a new method that systematically puts planning, motion prediction and safety metric design together to achieve environmental adaptive and safe navigation. This algorithm balances optimality in travel distance and safety with respect to passing clearance. Robot adapts progress speed adaptively according to the sensed environment, being fast in wide open areas and slow down in narrow passages and taking necessary maneuvers to avoid dangerous incoming obstacles. In our method, directional distance measure, conic-shape motion prediction and custom costmap are integrated properly to evaluate system risk accurately with respect to local geometry of surrounding environments. Using such risk estimation, reference governor technique and control barrier function are worked together to enable adaptive and safe path tracking in dynamical environments. We validate our algorithm extensively both in simulation and challenging real-world environments.

Deep Planar Parallax for Monocular Depth Estimation

Depth estimation is a fundamental problem in the perception system of autonomous driving scenes. Although autonomous driving is challenging, much prior knowledge can still be utilized, by which the sophistication of the problem can be effectively restricted. Some previous works introduce the road plane prior to the depth estimation problem according to the Planar Parallax Geometry. However, we find that their usages are not effective, leaving the network cannot learn the geometric information. To this end, we analyze this problem in detail and reveal that explicit warping of consecutive frames and flow pre-training can effectively bring the geometric prior into learning. Furthermore, we propose Planar Position Embedding to deal with the intrinsic weakness of plane parallax geometry. Comprehensive experimental results on autonomous driving datasets like KITTI and Waymo Open Dataset (WOD) demonstrate that our Planar Parallax Network(PPNet) dramatically outperforms existing learning-based methods.

Robust and Safe Autonomous Navigation for Systems with Learned SE(3) Hamiltonian Dynamics

Stability and safety are critical properties for successful deployment of automatic control systems. As a motivating example, consider autonomous mobile robot navigation in a complex environment. A control design that generalizes to different operational conditions requires a model of the system dynamics, robustness to modeling errors, and satisfaction of safety \NEWZL{constraints}, such as collision avoidance. This paper develops a neural ordinary differential equation network to learn the dynamics of a Hamiltonian system from trajectory data. The learned Hamiltonian model is used to synthesize an energy-shaping passivity-based controller and analyze its \emph{robustness} to uncertainty in the learned model and its \emph{safety} with respect to constraints imposed by the environment. Given a desired reference path for the system, we extend our design using a virtual reference governor to achieve tracking control. The governor state serves as a regulation point that moves along the reference path adaptively, balancing the system energy level, model uncertainty bounds, and distance to safety violation to guarantee robustness and safety. Our Hamiltonian dynamics learning and tracking control techniques are demonstrated on \Revised{simulated hexarotor and quadrotor robots} navigating in cluttered 3D environments.

Prediction of GPU Failures Under Deep Learning Workloads

Graphics processing units (GPUs) are the de facto standard for processing deep learning (DL) tasks. Meanwhile, GPU failures, which are inevitable, cause severe consequences in DL tasks: they disrupt distributed trainings, crash inference services, and result in service level agreement violations. To mitigate the problem caused by GPU failures, we propose to predict failures by using ML models. This paper is the first to study prediction models of GPU failures under large-scale production deep learning workloads. As a starting point, we evaluate classic prediction models and observe that predictions of these models are both inaccurate and unstable. To improve the precision and stability of predictions, we propose several techniques, including parallel and cascade model-ensemble mechanisms and a sliding training method. We evaluate the performances of our various techniques on a four-month production dataset including 350 million entries. The results show that our proposed techniques improve the prediction precision from 46.3\% to 84.0\%.

Safe Autonomous Navigation for Systems with Learned SE(3) Hamiltonian Dynamics

Safe autonomous navigation in unknown environments is an important problem for ground, aerial, and underwater robots. This paper proposes techniques to learn the dynamics models of a mobile robot from trajectory data and synthesize a tracking controller with safety and stability guarantees. The state of a mobile robot usually contains its position, orientation, and generalized velocity and satisfies Hamilton's equations of motion. Instead of a hand-derived dynamics model, we use a dataset of state-control trajectories to train a translation-equivariant nonlinear Hamiltonian model represented as a neural ordinary differential equation (ODE) network. The learned Hamiltonian model is used to synthesize an energy-shaping passivity-based controller and derive conditions which guarantee safe regulation to a desired reference pose. Finally, we enable adaptive tracking of a desired path, subject to safety constraints obtained from obstacle distance measurements. The trade-off between the system's energy level and the distance to safety constraint violation is used to adaptively govern the reference pose along the desired path. Our safe adaptive controller is demonstrated on a simulated hexarotor robot navigating in unknown complex environments.

SimpleTrack: Understanding and Rethinking 3D Multi-object Tracking

3D multi-object tracking (MOT) has witnessed numerous novel benchmarks and approaches in recent years, especially those under the "tracking-by-detection" paradigm. Despite their progress and usefulness, an in-depth analysis of their strengths and weaknesses is not yet available. In this paper, we summarize current 3D MOT methods into a unified framework by decomposing them into four constituent parts: pre-processing of detection, association, motion model, and life cycle management. We then ascribe the failure cases of existing algorithms to each component and investigate them in detail. Based on the analyses, we propose corresponding improvements which lead to a strong yet simple baseline: SimpleTrack. Comprehensive experimental results on Waymo Open Dataset and nuScenes demonstrate that our final method could achieve new state-of-the-art results with minor modifications. Furthermore, we take additional steps and rethink whether current benchmarks authentically reflect the ability of algorithms for real-world challenges. We delve into the details of existing benchmarks and find some intriguing facts. Finally, we analyze the distribution and causes of remaining failures in \name\ and propose future directions for 3D MOT. Our code is available at https://github.com/TuSimple/SimpleTrack.

Serving DNN Models with Multi-Instance GPUs: A Case of the Reconfigurable Machine Scheduling Problem

Multi-Instance GPU (MIG) is a new feature introduced by NVIDIA A100 GPUs that partitions one physical GPU into multiple GPU instances. With MIG, A100 can be the most cost-efficient GPU ever for serving Deep Neural Networks (DNNs). However, discovering the most efficient GPU partitions is challenging. The underlying problem is NP-hard; moreover, it is a new abstract problem, which we define as the Reconfigurable Machine Scheduling Problem (RMS). This paper studies serving DNNs with MIG, a new case of RMS. We further propose a solution, MIG-serving. MIG- serving is an algorithm pipeline that blends a variety of newly designed algorithms and customized classic algorithms, including a heuristic greedy algorithm, Genetic Algorithm (GA), and Monte Carlo Tree Search algorithm (MCTS). We implement MIG-serving on Kubernetes. Our experiments show that compared to using A100 as-is, MIG-serving can save up to 40% of GPUs while providing the same throughput.

Comparison between safety methods control barrier function vs. reachability analysis

This report aims to compare two safety methods: control barrier function and Hamilton-Jacobi reachability analysis. We will consider the difference with a focus on the following aspects: generality of system dynamics, difficulty of construction and computation cost. A standard Dubins car model will be evaluated numerically to make the comparison more concrete.

Unsupervised Scale-consistent Depth Learning from Video

We propose a monocular depth estimator SC-Depth, which requires only unlabelled videos for training and enables the scale-consistent prediction at inference time. Our contributions include: (i) we propose a geometry consistency loss, which penalizes the inconsistency of predicted depths between adjacent views; (ii) we propose a self-discovered mask to automatically localize moving objects that violate the underlying static scene assumption and cause noisy signals during training; (iii) we demonstrate the efficacy of each component with a detailed ablation study and show high-quality depth estimation results in both KITTI and NYUv2 datasets. Moreover, thanks to the capability of scale-consistent prediction, we show that our monocular-trained deep networks are readily integrated into the ORB-SLAM2 system for more robust and accurate tracking. The proposed hybrid Pseudo-RGBD SLAM shows compelling results in KITTI, and it generalizes well to the KAIST dataset without additional training. Finally, we provide several demos for qualitative evaluation.