As machine learning models become increasingly prevalent in motion forecasting for autonomous vehicles (AVs), it is critical to ensure that model predictions are safe and reliable. In this paper, we examine the robustness of motion forecasting to non-causal perturbations. We construct a new benchmark for evaluating and improving model robustness by applying perturbations to existing data. Specifically, we conduct an extensive labeling effort to identify causal agents, or agents whose presence influences human drivers’ behavior, in the Waymo Open Motion Dataset (WOMD), and we use these labels to perturb the data by deleting non-causal agents from the scene. We evaluate a diverse set of state-of-the-art deep-learning models on our proposed benchmark and find that all evaluated models exhibit large shifts under non-causal perturbation: we observe a surprising 25-38% relative change in minADE as compared to the original. In addition, we investigate techniques to improve model robustness, including increasing the training dataset size and using targeted data augmentations that randomly drop non-causal agents throughout training. Finally, we release the causal agent labels as an extension to WOMD and the robustness benchmarks to aid the community in building more reliable and safe deep-learning models for motion forecasting1.

We present MotionDiffuser, a diffusion based representation for the joint distribution of future trajectories over multiple agents. Such representation has several key advantages: first, our model learns a highly multimodal distribution that captures diverse future outcomes. Second, the simple predictor design requires only a single L2 loss training objective, and does not depend on trajectory anchors. Third, our model is capable of learning the joint distribution for the motion of multiple agents in a permutation-invariant manner. Furthermore, we utilize a compressed trajectory representation via PCA, which improves model performance and allows for efficient computation of the exact sample log probability. Subsequently, we propose a general constrained sampling framework that enables controlled trajectory sampling based on differentiable cost functions. This strategy enables a host of applications such as enforcing rules and physical priors, or creating tailored simulation scenarios. MotionDiffuser can be combined with existing backbone architectures to achieve top motion forecasting results. We obtain state-of-the-art results for multi-agent motion prediction on the Waymo Open Motion Dataset.

Reliable forecasting of the future behavior of road agents is a critical component to safe planning in autonomous vehicles. Here, we represent continuous trajectories as sequences of discrete motion tokens and cast multi-agent motion prediction as a language modeling task over this domain. Our model, MotionLM, provides several advantages: First, it does not require anchors or explicit latent variable optimization to learn multimodal distributions. Instead, we leverage a single standard language modeling objective, maximizing the average log probability over sequence tokens. Second, our approach bypasses post-hoc interaction heuristics where individual agent trajectory generation is conducted prior to interactive scoring. Instead, MotionLM produces joint distributions over interactive agent futures in a single autoregressive decoding process. In addition, the model’s sequential factorization enables temporally causal conditional rollouts. The proposed approach establishes new state-of-the-art performance for multi-agent motion prediction on the Waymo Open Motion Dataset, ranking 1ston the interactive challenge leaderboard.

Motion forecasting for autonomous driving is a challenging task because complex driving scenarios involve a heterogeneous mix of static and dynamic inputs. It is an open problem how best to represent and fuse information about road geometry, lane connectivity, time-varying traffic light state, and history of a dynamic set of agents and their interactions into an effective encoding. To model this diverse set of input features, many approaches proposed to design an equally complex system with a diverse set of modality specific modules. This results in systems that are difficult to scale, extend, or tune in rigorous ways to trade off quality and efficiency. In this paper, we present Wayformer, a family of simple and homogeneous attention based architectures for motion forecasting. Wayformer offers a compact model description consisting of an attention based scene encoder and a decoder. In the scene encoder we study the choice of early, late and hierarchical fusion of input modalities. For each fusion type we explore strategies to trade off efficiency and quality via factorized attention or latent query attention. We show that early fusion, despite its simplicity, is not only modality agnostic but also achieves state-of-the-art results on both Waymo Open Motion Dataset (WOMD) and Argoverse leaderboards, demonstrating the effectiveness of our design philosophy.

Imitation learning (IL) is a simple and powerful way to use high-quality human driving data, which can be collected at scale, to produce human-like behavior. However, policies based on imitation learning alone often fail to sufficiently account for safety and reliability concerns. In this paper, we show how imitation learning combined with reinforcement learning using simple rewards can substan-tially improve the safety and reliability of driving policies over those learned from imitation alone. In particular, we train a policy on over lOOk miles of urban driving data, and measure its effectiveness in test scenarios grouped by different levels of collision likelihood. Our analysis shows that while imitation can perform well in low-difficulty scenarios that are well-covered by the demonstration data, our proposed approach significantly improves robustness on the most challenging scenarios (over 38 % reduction in failures). To our knowledge, this is the first application of a combined imitation and reinforcement learning approach in autonomous driving that utilizes large amounts of real- world human driving data.

Simulation is an essential tool to develop and benchmark autonomous vehicle planning software in a safe and cost-effective manner. However, realistic simulation requires accurate modeling of multi-agent interactive behaviors to be trustworthy, behaviors which can be highly nuanced and complex. To address these challenges, we introduce Waymax, a new data-driven simulator for autonomous driving in multi-agent scenes, designed for large-scale simulation and testing. Waymax uses publicly-released, real-world driving data (e.g., the Waymo Open Motion Dataset) to initialize or play back a diverse set of multi-agent simulated scenarios. It runs entirely on hardware accelerators such as TPUs/GPUs and supports in-graph simulation for training, making it suitable for modern large-scale, distributed machine learning workflows. To support online training and evaluation, Waymax includes several learned and hard-coded behavior models that allow for realistic interaction within simulation. To supplement Waymax, we benchmark a suite of popular imitation and reinforcement learning algorithms with ablation studies on different design decisions, where we highlight the effectiveness of routes as guidance for planning agents and the ability of RL to overfit against simulated agents.

We introduce a motion forecasting (behavior prediction) method that meets the latency requirements for autonomous driving in dense urban environments without sacrificing accuracy. A whole-scene sparse input representation allows StopNet to scale to predicting trajectories for hundreds of road agents with reliable latency. In addition to predicting trajectories, our scene encoder lends itself to predicting whole-scene probabilistic occupancy grids, a complementary output representation suitable for busy urban environments. Occupancy grids allow the AV to reason collectively about the behavior of groups of agents without processing their individual trajectories. We demonstrate the effectiveness of our sparse input representation and our model in terms of computation and accuracy over three datasets. We further show that co-training consistent trajectory and occupancy predictions improves upon state-of-the-art performance under standard metrics.

Behavior prediction models have proliferated in recent years, especially in the popular real-world robotics application of autonomous driving, where representing the distribution over possible futures of moving agents is essential for safe and comfortable motion planning. In these models, the choice of coordinate frames to represent inputs and outputs has crucial trade offs which broadly fall into one of two categories. Agent-centric models transform inputs and perform inference in agent-centric coordinates. These models are intrinsically invari-ant to translation and rotation between scene elements, are best-performing on public leaderboards, but scale quadratically with the number of agents and scene elements. Scene-centric models use a fixed coordinate system to process all agents. This gives them the advantage of sharing representations among all agents, offering efficient amortized inference computation which scales linearly with the number of agents. However, these models have to learn invariance to translation and rotation between scene elements, and typically underperform agent-centric models. In this work, we develop knowledge distillation techniques between probabilistic motion forecasting models, and apply these techniques to close the gap in performance between agent-centric and scene-centric models. This improves scene-centric model performance by 13.2% on the public Argoverse benchmark, 7.8% on Waymo Open Dataset and up to 9.4% on a large In-House dataset. These improved scene-centric models rank highly in public leaderboards and are up to 15 times more efficient than their agent-centric teacher counterparts in busy scenes.

Predicting the future behavior of road users is one of the most challenging and important problems in autonomous driving. Applying deep learning to this problem requires fusing heterogeneous world state in the form of rich perception signals and map information, and inferring highly multi-modal distributions over possible futures. In this paper, we present MultiPath++, a future prediction model that achieves state-of-the-art performance on popular benchmarks. MultiPath++ improves the MultiPath architecture [34] by revisiting many design choices. The first key design difference is a departure from dense image-based encoding of the input world state in favor of a sparse encoding of heterogeneous scene elements: MultiPath++ consumes compact and efficient polylines to describe road features, and raw agent state information directly (e.g., position, velocity, acceleration). We propose a context-aware fusion of these elements and develop a reusable multi-context gating fusion component. Second, we reconsider the choice of pre-defined static anchors, and develop a way to learn latent anchor embeddings end-to-end in the model. Lastly, we explore ensembling and output aggregation techniques—common in other ML domains—and find effective variants for our probabilistic multimodal output representation. We perform an extensive ablation on these design choices, and show that our proposed model achieves state-of-the-art performance on the Argoverse Motion Forecasting Competition [10] and the Waymo Open Dataset Motion Prediction Challenge [13].

As autonomous driving systems mature, motion forecasting has received increasing attention as a critical requirement for planning. Of particular importance are interactive situations such as merges, unprotected turns, etc., where predicting individual object motion is not sufficient. Joint predictions of multiple objects are required for effective route planning. There has been a critical need for high-quality motion data that is rich in both interactions and annotation to develop motion planning models. In this work, we introduce the most diverse interactive motion dataset to our knowledge, and provide specific labels for interacting objects suitable for developing joint prediction models. With over 100,000 scenes, each 20 seconds long at 10 Hz, our new dataset contains more than 570 hours of unique data over 1750 km of roadways. It was collected by mining for interesting interactions between vehicles, pedestrians, and cyclists across six cities within the United States. We use a high-accuracy 3D auto-labeling system to generate high quality 3D bounding boxes for each road agent, and provide corresponding high definition 3D maps for each scene. Furthermore, we introduce a new set of metrics that provides a comprehensive evaluation of both single agent and joint agent interaction motion forecasting models. Finally, we provide strong baseline models for individual-agent prediction and joint-prediction. We hope that this new large-scale interactive motion dataset will provide new opportunities for advancing motion forecasting models.

Interactive driving scenarios, such as lane changes, merges and unprotected turns, are some of the most challenging situations for autonomous driving. Planning in interactive scenarios requires accurately modeling the reactions of other agents to different future actions of the ego agent. We develop end-to-end models for conditional behavior prediction (CBP) that take as an input a query future trajectory for an ego-agent, and predict distributions over future trajectories for other agents conditioned on the query. Leveraging such a model, we develop a general-purpose agent interactivity score derived from probabilistic first principles. The interactivity score allows us to find interesting interactive scenarios for training and evaluating behavior prediction models. We further demonstrate that the proposed score is effective for agent prioritization under computational budget constraints.

We focus on the problem of predicting future states of entities in complex, real-world driving scenarios. Previous research has approached this problem via low-level signals to predict short time horizons, and has not addressed how to leverage key assets relied upon heavily by industry self-driving systems: (1) large 3D perception efforts which provide highly accurate 3D states of agents with rich attributes, and (2) detailed and accurate semantic maps of the environment (lanes, traffic lights, crosswalks, etc). We present a unified representation which encodes such high-level semantic information in a spatial grid, allowing the use of deep convolutional models to fuse complex scene context. This enables learning entity-entity and entity-environment interactions with simple, feed-forward computations in each timestep within an overall temporal model of an agent's behavior. We propose different ways of modelling the future as a {\em distribution} over future states using standard supervised learning. We introduce a novel dataset providing industry-grade rich perception and semantic inputs, and empirically show we can effectively learn fundamentals of driving behavior.

We propose a multimodal, decomposable model for articulated human pose estimation in monocular images. A typical approach to this problem is to use a linear structured model, which struggles to capture the wide range of appearance present in realistic, unconstrained images. In this paper, we instead propose a model of human pose that explicitly captures a variety of pose modes. Unlike other multimodal models, our approach includes both global and local pose cues and uses a convex objective and joint training for mode selection and pose estimation. We also employ a cascaded mode selection step which controls the trade-off between speed and accuracy, yielding a 5x speedup in inference and learning. Our model outperforms state-of-the-art approaches across the accuracy-speed trade-off curve for several pose datasets. This includes our newly-collected dataset of people in movies, FLIC, which contains an order of magnitude more labeled data for training and testing than existing datasets.

In many cases, the predictive power of structured models for for complex vision tasks is limited by a trade-off between the expressiveness and the computational tractability of the model. However, choosing this trade-off statically a priori is sub optimal, as images and videos in different settings vary tremendously in complexity. On the other hand, choosing the trade-off dynamically requires knowledge about the accuracy of different structured models on any given example. In this work, we propose a novel two-tier architecture that provides dynamic speed/accuracy trade-offs through a simple type of introspection. Our approach, which we call dynamic structured model selection (DMS), leverages typically intractable features in structured learning problems in order to automatically determine' which of several models should be used at test-time in order to maximize accuracy under a fixed budgetary constraint. We demonstrate DMS on two sequential modeling vision tasks, and we establish a new state-of-the-art in human pose estimation in video with an implementation that is roughly 23× faster than the previous standard implementation.

Modern smartphones are equipped with a plethora of sensors that enable a wide range of interactions, but some of these sensors can be employed as a side channel to surreptitiously learn about user input. In this paper, we show that the accelerometer sensor can also be employed as a high-bandwidth side channel; particularly, we demonstrate how to use the accelerometer sensor to learn user tap- and gesture-based input as required to unlock smartphones using a PIN/password or Android's graphical password pattern. Using data collected from a diverse group of 24 users in controlled (while sitting) and uncontrolled (while walking) settings, we develop sample rate independent features for accelerometer readings based on signal processing and polynomial fitting techniques. In controlled settings, our prediction model can on average classify the PIN entered 43% of the time and pattern 73% of the time within 5 attempts when selecting from a test set of 50 PINs and 50 patterns. In uncontrolled settings, while users are walking, our model can still classify 20% of the PINs and 40% of the patterns within 5 attempts. We additionally explore the possibility of constructing an accelerometer-reading-to-input dictionary and find that such dictionaries would be greatly challenged by movement-noise and cross-user training.

We address the problem of articulated human pose estimation in videos using an ensemble of tractable models with rich appearance, shape, contour and motion cues. In previous articulated pose estimation work on unconstrained videos, using temporal coupling of limb positions has made little to no difference in performance over parsing frames individually. One crucial reason for this is that joint parsing of multiple articulated parts over time involves intractable inference and learning problems, and previous work has resorted to approximate inference and simplified models. We overcome these computational and modeling limitations using an ensemble of tractable submodels which couple locations of body joints within and across frames using expressive cues. Each submodel is responsible for tracking a single joint through time (e.g., left elbow) and also models the spatial arrangement of all joints in a single frame. Because of the tree structure of each submodel, we can perform efficient exact inference and use rich temporal features that depend on image appearance, e.g., color tracking and optical flow contours. We propose and experimentally investigate a hierarchy of submodel combination methods, and we find that a highly efficient max-marginal combination method outperforms much slower (by orders of magnitude) approximate inference using dual decomposition. We apply our pose model on a new video dataset of highly varied and articulated poses from TV shows. We show significant quantitative and qualitative improvements over state-of-the-art single-frame pose estimation approaches.

We address the character identification problem in movies and television videos: assigning names to faces on the screen. Most prior work on person recognition in video assumes some supervised data such as screenplay or handlabeled faces. In this paper, our only source of `supervision' are the dialog cues: first, second and third person references (such as “I'm Jack”, “Hey, Jack!” and “Jack left”). While this kind of supervision is sparse and indirect, we exploit multiple modalities and their interactions (appearance, dialog, mouth movement, synchrony, continuity-editing cues) to effectively resolve identities through local temporal grouping followed by global weakly supervised recognition. We propose a novel temporal grouping model that partitions face tracks across multiple shots while respecting appearance, geometric and film-editing cues and constraints. In this model, states represent partitions of the k most recent face tracks, and transitions represent compatibility of consecutive partitions. We present dynamic programming inference and discriminative learning for the model. The individual face tracks are subsequently assigned a name by learning a classifier from partial label constraints. The weakly supervised classifier incorporates multiple-instance constraints from dialog cues as well as soft grouping constraints from our temporal grouping. We evaluate both the temporal grouping and final character naming on several hours of TV and movies.

Pictorial structure (PS) models are extensively used for part-based recognition of scenes, people, animals and multi-part objects. To achieve tractability, the structure and parameterization of the model is often restricted, for example, by assuming tree dependency structure and unimodal, data-independent pairwise interactions. These expressivity restrictions fail to capture important patterns in the data. On the other hand, local methods such as nearest-neighbor classification and kernel density estimation provide non-parametric flexibility but require large amounts of data to generalize well. We propose a simple semi-parametric approach that combines the tractability of pictorial structure inference with the flexibility of non-parametric methods by expressing a subset of model parameters as kernel regression estimates from a learned sparse set of exemplars. This yields query-specific, image-dependent pose priors. We develop an effective shape-based kernel for upper-body pose similarity and propose a leave-one-out loss function for learning a sparse subset of exemplars for kernel regression. We apply our techniques to two challenging datasets of human figure parsing and advance the state-of-the-art (from 80% to 86% on the Buffy dataset), while using only 15% of the training data as exemplars.

For many structured prediction problems, complex models often require adopting approximate inference techniques such as variational methods or sampling, which generally provide no satisfactory accuracy guarantees. In this work, we propose sidestepping intractable inference altogether by learning ensembles of tractable sub-models as part of a structured prediction cascade. We focus in particular on problems with high-treewidth and large state-spaces, which occur in many computer vision tasks. Unlike other variational methods, our ensembles do not enforce agreement between sub-models, but filter the space of possible outputs by simply adding and thresholding the max-marginals of each constituent model. Our framework jointly estimates parameters for all models in the ensemble for each level of the cascade by minimizing a novel, convex loss function, yet requires only a linear increase in computation over learning or inference in a single tractable sub-model. We provide a generalization bound on the filtering loss of the ensemble as a theoretical justification of our approach, and we evaluate our method on both synthetic data and the task of estimating articulated human pose from challenging videos. We find that our approach significantly outperforms loopy belief propagation on the synthetic data and a state-of-the-art model on the pose estimation/tracking problem.

In many image and video collections, we have access only to partially labeled data. For example, personal photo collections often contain several faces per image and a caption that only specifies who is in the picture, but not which name matches which face. Similarly, movie screenplays can tell us who is in the scene, but not when and where they are on the screen. We formulate the learning problem in this setting as partially-supervised multiclass classification where each instance is labeled ambiguously with more than one label. We show theoretically that effective learning is possible under reasonable assumptions even when all the data is weakly labeled. Motivated by the analysis, we propose a general convex learning formulation based on minimization of a surrogate loss appropriate for the ambiguous label setting. We apply our framework to identifying faces culled from Web news sources and to naming characters in TV series and movies. We experiment on a very large dataset consisting of 100 hours of video, and in particular achieve 6% error for character naming on 16 episodes of LOST.