Scene flow estimation is a crucial component in the development of autonomous driving and 3D robotics, providing valuable information for environment perception and navigation. Despite the advantages of learning-based scene flow estimation techniques, their domain specificity and limited generalizability across varied scenarios pose challenges. In contrast, non-learning optimization-based methods, incorporating robust priors or regularization, offer competitive scene flow estimation performance, require no training, and show extensive applicability across datasets, but suffer from lengthy inference times.In this paper, we present OptFlow, a fast optimization-based scene flow estimation method. Without relying on learning or any labeled datasets, OptFlow achieves state-of-the-art performance for scene flow estimation on popular autonomous driving benchmarks. It integrates a local correlation weight matrix for correspondence matching, an adaptive correspondence threshold limit for nearest-neighbor search, and graph prior rigidity constraints, resulting in expedited convergence and improved point correspondence identification. Moreover, we demonstrate how integrating a point cloud registration function within our objective function bolsters accuracy and differentiates between static and dynamic points without relying on external odometry data. Consequently, OptFlow outperforms the baseline graph-prior method by approximately 20% and the Neural Scene Flow Prior method by 5%-7% in accuracy, all while offering the fastest inference time among all nonlearning scene flow estimation methods.

Simulation has the potential to transform the development of robust algorithms for mobile agents deployed in safety-critical scenarios. However, the poor photorealism and lack of diverse sensor modalities of existing simulation engines remain key hurdles towards realizing this potential. Here, we present VISTA††Full code release for the VISTA data-driven simulation engine is available here: vista.csail.mit.edu., an open source, data-driven simulator that integrates multiple types of sensors for autonomous vehicles. Using high fidelity, real-world datasets, VISTA represents and simulates RGB cameras, 3D LiDAR, and event-based cameras, enabling the rapid generation of novel viewpoints in simulation and thereby enriching the data available for policy learning with corner cases that are difficult to capture in the physical world. Using VISTA, we demonstrate the ability to train and test perception-to-control policies across each of the sensor types and showcase the power of this approach via deployment on a full scale autonomous vehicle. The policies learned in VISTA exhibit sim-to-real transfer without modification and greater robustness than those trained exclusively on real-world data.

Data-driven simulators promise high data-efficiency for driving policy learning. When used for modelling interactions, this data-efficiency becomes a bottleneck: small underlying datasets often lack interesting and challenging edge cases for learning interactive driving. We address this challenge by proposing a data-driven simulation engine† that uses inpainted ado vehicles for learning robust driving policies. Thus, our approach can be used to learn policies that involve multi-agent interactions and allows for training via state-of-the-art policy learning methods. We evaluate the approach for learning standard interaction scenarios in driving. In extensive experiments, our work demonstrates that the resulting policies can be directly transferred to a full-scale autonomous vehicle without making use of any traditional sim-to-real transfer techniques such as domain randomization.

Autonomous control of an emergency vehicle will save lives through faster transport and shorter response. Towards this goal, it must overcome the challenge of inter- acting with existing human drivers on the road. We present a game-theoretic approach for semi-cooperative control of an autonomous emergency vehicle that can interact efficiently with humans on the road. We model the interactions between autonomous and human driven cars with Social Value Orientation, a metric from social psychology, that allows the controller to leverage their influence on the trajectories of neighboring human drivers. In addition, by using a modified version of iterative best response, we direct the algorithm to converge to Nash equilibria that are cooperative. We demonstrate the efficacy of our algorithm in simulations of drivers in traffic, with a variety of traffic densities and driver personalities. In simulations of prosocial human drivers, our algorithm provides an 8% improvement in distance-traveled compared to egoistic human drivers.

Reinforcement learning (RL) for continuous control typically employs distributions whose support covers the entire action space. In this work, we investigate the colloquially known phenomenon that trained agents often prefer actions at the boundaries of that space. We draw theoretical connections to the emergence of bang-bang behavior in optimal control, and provide extensive empirical evaluation across a variety of recent RL algorithms. We replace the normal Gaussian by a Bernoulli distribution that solely considers the extremes along each action dimension - a bang-bang controller. Surprisingly, this achieves state-of-the-art performance on several continuous control benchmarks - in contrast to robotic hardware, where energy and maintenance cost affect controller choices. Since exploration, learning, and the final solution are entangled in RL, we provide additional imitation learning experiments to reduce the impact of exploration on our analysis. Finally, we show that our observations generalize to environments that aim to model real-world challenges and evaluate factors to mitigate the emergence of bang-bang solutions. Our findings emphasise challenges for benchmarking continuous control algorithms, particularly in light of potential real-world applications.

Information gathering while interacting with other agents under sensing and motion uncertainty is critical in domains such as driving, service robots, racing, or surveillance. The interests of agents may be at odds with others, resulting in a stochastic noncooperative dynamic game. Agents must predict others’ future actions without communication, incorporate their actions into these predictions, account for uncertainty and noise in information gathering, and consider what information their actions reveal. Our solution uses local iterative dynamic programming in Gaussian belief space to solve a game-theoretic continuous POMDP. Solving a quadratic game in the backward pass of a game-theoretic belief-space variant of iterative linear-quadratic Gaussian control (iLQG) achieves a runtime polynomial in the number of agents and linear in the planning horizon. Our algorithm yields linear feedback policies for our robot, and predicted feedback policies for other agents. We present three applications: Active surveillance, guiding eyes for a blind agent, and autonomous racing. Agents with game-theoretic belief-space planning win 44% more races than without game theory and 34% more than without belief-space planning.

This paper introduces the Weighted Buffered Voronoi tessellation, which allows us to define distributed, semicooperative multi-agent navigation policies with guarantees on collision avoidance. We generate the Voronoi cells with dynamic weights that bias the boundary towards the agent with the lower relative weight while always maintaining a buffered distance between two agents. By incorporating agent weights, we can encode selfish or prioritized behavior among agents, where a more selfish agent will have a larger relative cell over less selfish agents. We consider this semi-cooperative since agents do not cooperate in symmetric ways. Furthermore, when all agents start in a collision-free configuration and plan their control actions within their cells, we prove that no agents will collide. Simulations demonstrate the performance of our algorithm for agents navigating to goal locations in a position-swapping game. We observe that agents with more egoistic weights consistently travel shorter paths to their goal than more altruistic agents.

This paper investigates the safe path planning problem for an autonomous vehicle operating in unstructured, cluttered environments. While some objects may be accurately with canonical perception algorithms, other objects and clutter may be harder to track. We present an approach that combines two methods of risk assessment: for objects with reliable tracking, we use a Gaussian Process (GP) regulated risk map to describe the risk map information; for unknown objects that we fail to accurately track, we compute a Dynamic Risk Density (DRD) from the overall occupancy and velocity field from LiDAR scan snapshots. Several methods are proposed for combining the GP risk map and DRD, and the resultant hybrid risk map is used for the proposed safe path planning algorithm. Experimental results on an autonomous buggy show that the hybrid risk map is able to yield a safe path planner to navigate the autonomous testbed within the cluttered environments.

Deterministic neural networks (NNs) are increasingly being deployed in safety critical domains, where calibrated, robust, and efficient measures of uncertainty are crucial. In this paper, we propose a novel method for training non-Bayesian NNs to estimate a continuous target as well as its associated evidence in order to learn both aleatoric and epistemic uncertainty. We accomplish this by placing evidential priors over the original Gaussian likelihood function and training the NN to infer the hyperparameters of the evidential distribution. We additionally impose priors during training such that the model is regularized when its predicted evidence is not aligned with the correct output. Our method does not rely on sampling during inference or on out-of-distribution (OOD) examples for training, thus enabling efficient and scalable uncertainty learning. We demonstrate learning well-calibrated measures of uncertainty on various benchmarks, scaling to complex computer vision tasks, as well as robustness to adversarial and OOD test samples.

Recent research has highlighted the vulnerabilities of modern machine learning based systems to bias, especially towards segments of society that are under-represented in training data. In this work, we develop a novel, tunable algorithm for mitigating the hidden, and potentially unknown, biases within training data. Our algorithm fuses the original learning task with a variational autoencoder to learn the latent structure within the dataset and then adaptively uses the learned latent distributions to re-weight the importance of certain data points while training. While our method is generalizable across various data modalities and learning tasks, in this work we use our algorithm to address the issue of racial and gender bias in facial detection systems. We evaluate our algorithm on the Pilot Parliaments Benchmark (PPB), a dataset specifically designed to evaluate biases in computer vision systems, and demonstrate increased overall performance as well as decreased categorical bias with our debiasing approach.

In this paper, we examine the problem of navigating cluttered environments without explicit object detection and tracking. We introduce the dynamic risk density to map the congestion density and spatial flow of the environment to a cost function for the agent to determine risk when navigating that environment. We build upon our prior work, wherein the agent maps the density and motion of objects to an occupancy risk, then navigate the environment over a specified risk level set. Here, the agent does not need to identify objects to compute the occupancy risk, and instead computes this cost function using the occupancy density and velocity fields around them. Simulations show how this dynamic risk density encodes movement information for the ego agent and closely models the object-based congestion cost. We implement our dynamic risk density on an autonomous wheelchair and show how it can be used for navigating unstructured, crowded and cluttered environments.

Current methods for autonomous management use strict first-come, first-serve (FCFS) ordering to manage incoming autonomous vehicles at an intersection. In this work, we present a coordination policy that swaps agent ordering to increase the system-wide performance while ensuring that the swaps are socially compliant. By considering an agent's Social Value Orientation (SVO), a social psychology metric for their willingness to help another vehicle, the central coordinator can reduce system delays while ensuring each individual vehicle increases their own utility. The FCFS-SVO algorithm is both computationally tractable and accounts for a variety of real-world agent types, such as human drivers and a variety of social orientations. Simulation results show that average vehicle delays decrease with swapping by enabling cooperation between agents. In addition, we show that the proportion of human drivers, as well as, the distribution of prosocial and egoistic vehicles in the system can have a prominent effect on the performance of the system.

This paper examines how vehicles can quickly quantify the level of congestion in their environment for planning. We use risk level sets to define a metric of congestion for the vehicles. Using this metric, we can quickly identify distributions of environment and driver features, such as velocities and number of neighbors, based on risk within human driving data sets. We use the NGSIM and highD data sets to study how risk influences behaviors in city and highway driving. From these data sets, we learn common risk thresholds for classifying low, medium, and high-risk situations. Using these thresholds, we develop simulations of an autonomous vehicle driving along a highway, and demonstrate how the chosen risk threshold influences the autonomous vehicle behavior.

In this paper, we address the problem of navigating in a cluttered environment by introducing a congestion cost that maps the density and motion of objects to an occupancy risk. We propose that an agent can choose a “risk level set” from this cost function and construct a planning space from this set. In choosing different levels of risk, the agent adjusts its interactions with the other agents. From the assumption that agents are self-preserving, we show that any agent planning within their risk level set will avoid collisions with other agents. We then present an application of planning with risk level sets in the framework of an autonomous vehicle driving along a highway. Using the risk level sets, the agent can determine safe zones when planning a sequence of lane changes. Through simulations in Matlab, we demonstrate how the choice of risk threshold manifests as aggressive or conservative behavior.

When driving in urban environments, an autonomous vehicle must account for the interaction with other traffic participants. It must reason about their future behavior, how its actions affect their future behavior, and potentially consider multiple motion hypothesis. In this paper we introduce a method for joint behavior estimation and trajectory planning that models interaction and multi-policy decision-making. The method leverages Partially Observable Markov Decision Processes to estimate the behavior of other traffic participants given the planned trajectory for the ego-vehicle, and Receding-Horizon Control for generating safe trajectories for the ego-vehicle. To achieve safe navigation we introduce chance constraints over multiple motion policies in the receding-horizon planner. These constraints account for uncertainty over the behavior of other traffic participants. The method is capable of running in real-time and we show its performance and good scalability in simulated multi-vehicle intersection scenarios.

This paper introduces a new method for end-to-end training of deep neural networks (DNNs) and evaluates it in the context of autonomous driving. DNN training has been shown to result in high accuracy for perception to action learning given sufficient training data. However, the trained models may fail without warning in situations with insufficient or biased training data. In this paper, we propose and evaluate a novel architecture for self-supervised learning of latent variables to detect the insufficiently trained situations. Our method also addresses training data imbalance, by learning a set of underlying latent variables that characterize the training data and evaluate potential biases. We show how these latent distributions can be leveraged to adapt and accelerate the training pipeline by training on only a fraction of the total dataset. We evaluate our approach on a challenging dataset for driving. The data is collected from a full-scale autonomous vehicle. Our method provides qualitative explanation for the latent variables learned in the model. Finally, we show how our model can be additionally trained as an end-to-end controller, directly outputting a steering control command for an autonomous vehicle.

High-end vehicles are already equipped with safety systems, such as assistive braking and automatic lane following, enhancing vehicle safety. Yet, these current solutions can only help in low-complexity driving situations. In this paper, we introduce a parallel autonomy, or shared control, framework that computes safe trajectories for an automated vehicle, based on human inputs. We minimize the deviation from the human inputs while ensuring safety via a set of collision avoidance constraints. Our method achieves safe motion even in complex driving scenarios, such as those commonly encountered in an urban setting. We introduce a receding horizon planner formulated as nonlinear model predictive control (NMPC), which includes the analytic descriptions of road boundaries and the configuration and future uncertainties of other road participants. The NMPC operates over both steering and acceleration simultaneously. We introduce a nonslip model suitable for handling complex environments with dynamic obstacles, and a nonlinear combined slip vehicle model including normal load transfer capable of handling static environments. We validate the proposed approach in two complex driving scenarios. First, in an urban environment that includes a left-turn across traffic and passing on a busy street. And second, under snow conditions on a race track with sharp turns and under complex dynamic constraints. We evaluate the performance of the method with various human driving styles. We consequently observe that the method successfully avoids collisions and generates motions with minimal intervention for parallel autonomy. We note that the method can also be applied to generate safe motion for fully autonomous vehicles.

Current state-of-the-art vehicle safety systems, such as assistive braking or automatic lane following, are still only able to help in relatively simple driving situations. We introduce a Parallel Autonomy shared-control framework that produces safe trajectories based on human inputs even in much more complex driving scenarios, such as those commonly encountered in an urban setting. We minimize the deviation from the human inputs while ensuring safety via a set of collision avoidance constraints. We develop a receding horizon planner formulated as a Non-linear Model Predictive Control (NMPC) including analytic descriptions of road boundaries, and the configurations and future uncertainties of other traffic participants, and directly supplying them to the optimizer without linearization. The NMPC operates over both steering and acceleration simultaneously. Furthermore, the proposed receding horizon planner also applies to fully autonomous vehicles. We validate the proposed approach through simulations in a wide variety of complex driving scenarios such as left-turns across traffic, passing on busy streets, and under dynamic constraints in sharp turns on a race track.

Recent advances in autonomous driving have raised the problem of safety to the forefront and incentivized research into establishing safety guarantees. In this paper, we propose a safety verification framework as a safety standard for driving controllers with full or shared autonomy based on compositional and contract-based principles. Our framework enables us to synthesize safety guarantees over entire road networks by first building a library of locally verified models, and then composing local models together to verify the entire network. Composition is achieved using assume-guarantee contracts that are synthesized concurrently during verification. Thus, we can reuse local models within and across networks, add additional models to cover local road geometries without re-verifying the entire library, and perform all computations in a parallel and distributed way, which enables computational tractability. Furthermore, we employ controller contracts such that any controller satisfying them can be certified safe. We demonstrate the practical effectiveness of our framework by certifying controllers over parts of the Manhattan road network.

We present the development of a full-scale “parallel autonomy” research platform including software and hardware. In the parallel autonomy paradigm, the control of the vehicle is shared; the human is still in control of the vehicle, but the autonomy system is always running in the background to prevent accidents. Our holistic approach includes: (1) a drive-by-wire conversion method only based on reverse engineering mounting of relatively inexpensive sensors onto the vehicle implementation of a localization and mapping system, (4) obstacle detection and (5) a shared controller as well as (6) integration with an advanced autonomy simulation system (Drake) for rapid development and testing. The system can operate in three modes: (a) manual driving, (b) full autonomy, where the system is in complete control of the vehicle and (c) parallel autonomy, where the shared controller is implemented. We present results from extensive testing of a full-scale vehicle on closed tracks that demonstrate these capabilities.