A broad challenge of research on generalization for sequential decision-making tasks in interactive environments is designing benchmarks that clearly landmark progress. While there has been notable headway, current benchmarks either do not provide suitable exposure nor intuitive control of the underlying factors, are not easy-to-implement, customizable, or extensible, or are computationally expensive to run. We built the Sandbox Environment for Generalizable Agent Research (SEGAR) with all of these things in mind. SEGAR improves the ease and accountability of generalization research in RL, as generalization objectives can be easy designed by specifying task distributions, which in turns allows the researcher to measure the nature of the generalization objective. We present an overview of SEGAR and how it contributes to these goals, as well as experiments that demonstrate a few types of research questions SEGAR can help answer.

Evaluating autonomous vehicle stacks (AVs) in simulation typically involves replaying driving logs from real-world recorded traffic. However, agents replayed from offline data are not reactive and hard to intuitively control. Existing approaches address these challenges by proposing methods that rely on heuristics or generative models of real-world data but these approaches either lack realism or necessitate costly iterative sampling procedures to control the generated behaviours. In this work, we take an alternative approach and propose CtRL-Sim, a method that leverages return-conditioned offline reinforcement learning to efficiently generate reactive and controllable traffic agents. Specifically, we process real-world driving data through a physics-enhanced Nocturne simulator to generate a diverse offline reinforcement learning dataset, annotated with various reward terms. With this dataset, we train a return-conditioned multi-agent behaviour model that allows for fine-grained manipulation of agent behaviours by modifying the desired returns for the various reward components. This capability enables the generation of a wide range of driving behaviours beyond the scope of the initial dataset, including adversarial behaviours. We demonstrate that CtRL-Sim can generate diverse and realistic safety-critical scenarios while providing fine-grained control over agent behaviours.

The field of Continual Learning (CL) seeks to develop algorithms that accumulate knowledge and skills over time through interaction with non-stationary environments. In practice, a plethora of evaluation procedures (settings) and algorithmic solutions (methods) exist, each with their own potentially disjoint set of assumptions. This variety makes measuring progress in CL difficult. We propose a taxonomy of settings, where each setting is described as a set of assumptions. A tree-shaped hierarchy emerges from this view, where more general settings become the parents of those with more restrictive assumptions. This makes it possible to use inheritance to share and reuse research, as developing a method for a given setting also makes it directly applicable onto any of its children. We instantiate this idea as a publicly available software framework called Sequoia, which features a wide variety of settings from both the Continual Supervised Learning (CSL) and Continual Reinforcement Learning (CRL) domains. Sequoia also includes a growing suite of methods which are easy to extend and customize, in addition to more specialized methods from external libraries. We hope that this new paradigm and its first implementation can help unify and accelerate research in CL. You can help us grow the tree by visiting www.github.com/lebrice/Sequoia.

Different types of mental rotation tests have been used extensively in psychology to understand human visual reasoning and perception. Understanding what an object or visual scene would look like from another viewpoint is a challenging problem that is made even harder if it must be performed from a single image. We explore a controlled setting whereby questions are posed about the properties of a scene if that scene was observed from another viewpoint. To do this we have created a new version of the CLEVR dataset that we call CLEVR Mental Rotation Tests (CLEVR-MRT). Using CLEVR-MRT we examine standard methods, show how they fall short, then explore novel neural architectures that involve inferring volumetric representations of a scene. These volumes can be manipulated via camera-conditioned transformations to answer the question. We examine the efficacy of different model variants through rigorous ablations and demonstrate the efficacy of volumetric representations.

We consider the problem of estimating an object's physical properties such as mass, friction, and elasticity directly from video sequences. Such a system identification problem is fundamentally ill-posed due to the loss of information during image formation. Current solutions require precise 3D labels which are labor-intensive to gather, and infeasible to create for many systems such as deformable solids or cloth. We present gradSim, a framework that overcomes the dependence on 3D supervision by leveraging differentiable multiphysics simulation and differentiable rendering to jointly model the evolution of scene dynamics and image formation. This novel combination enables backpropagation from pixels in a video sequence through to the underlying physical attributes that generated them. Moreover, our unified computation graph -- spanning from the dynamics and through the rendering process -- enables learning in challenging visuomotor control tasks, without relying on state-based (3D) supervision, while obtaining performance competitive to or better than techniques that rely on precise 3D labels.

Abstract--Robots in many real-world settings have access to force/torque sensors in their gripper and tactile sensing is often necessary in tasks that involve contact-rich motion. In this work, we leverage surprise from mismatches in touch feedback to guide exploration in hard sparse-reward reinforcement learning tasks. Our approach, Touch-based Curiosity (ToC), learns what visible objects interactions are supposed to "feel" like. We encourage exploration by rewarding interactions where the expectation and the experience don't match. In our proposed method, an initial task-independent exploration phase is followed by an on-task learning phase, in which the original interactions are relabeled with on-task rewards. We test our approach on a range of touchintensive robot arm tasks (e.g. In the former, the environment is often fully observable, and the reward is dense and well-defined. In the Recent works in RL have focused on curiosity-driven latter, a large amount of work is required to design useful exploration through prediction-based surprise [6, 45, 48]. While it may be possible to hand-craft dense formulation, a forward dynamics models predicts the future, and reward signals for many real-world tasks, we believe that it's if its prediction is incorrect when compared to the real future, a worthwhile endeavor to investigate learning methods that do the agent is surprised and is thus rewarded.

Humans have the innate ability to attend to the most relevant actors in their vicinity and can forecast how they may behave in the future. This ability will be crucial for the deployment of safety-critical agents such as robots or vehicles which interact with humans. We propose a theoretical framework for this problem setting based on autoregressively modelling sequences of nested sets, using latent variables to better capture multimodal distributions over future sets of sets. We present a new model architecture which we call a Nested Set Transformer which employs multi-head self-attention blocks over sets of sets that serve as a form of social attention between the elements of the sets at every timestep. Our approach can produce a distribution over future trajectories for all agents under consideration, or focus upon the trajectory of an ego-agent. We validate the Nested Set Transformer for autonomous driving settings which we refer to as ("AutoBot"), where we model the trajectory of an ego-agent based on the sequential observations of key attributes of multiple agents in a scene. AutoBot produces results better than state-of-the-art published prior work on the challenging nuScenes vehicle trajectory modeling benchmark. We also examine the multi-agent prediction version of our model and jointly forecast an ego-agent's future trajectory along with the other agents in the scene. We validate the behavior of our proposed Nested Set Transformer for scene level forecasting with a pedestrian trajectory dataset.

This report presents the debates, posters, and discussions of the Sim2Real workshop held in conjunction with the 2020 edition of the "Robotics: Science and System" conference. Twelve leaders of the field took competing debate positions on the definition, viability, and importance of transferring skills from simulation to the real world in the context of robotics problems. The debaters also joined a large panel discussion, answering audience questions and outlining the future of Sim2Real in robotics. Furthermore, we invited extended abstracts to this workshop which are summarized in this report. Based on the workshop, this report concludes with directions for practitioners exploiting this technology and for researchers further exploring open problems in this area.

Domain randomization is a popular technique for improving domain transfer, often used in a zero-shot setting when the target domain is unknown or cannot easily be used for training. In this work, we empirically examine the effects of domain randomization on agent generalization. Our experiments show that domain randomization may lead to suboptimal, high-variance policies, which we attribute to the uniform sampling of environment parameters. We propose Active Domain Randomization, a novel algorithm that learns a parameter sampling strategy. Our method looks for the most informative environment variations within the given randomization ranges by leveraging the discrepancies of policy rollouts in randomized and reference environment instances. We find that training more frequently on these instances leads to better overall agent generalization. In addition, when domain randomization and policy transfer fail, Active Domain Randomization offers more insight into the deficiencies of both the chosen parameter ranges and the learned policy, allowing for more focused debugging. Our experiments across various physics-based simulated and a real-robot task show that this enhancement leads to more robust, consistent policies.