Typical robotic systems rely on models for planning. Therefore, the quality of the robot's behavior is heavily dependent on how accurately the model can predict the outcome of the robot's actions in the environment. A challenge, however, is that no model is perfect; moreover, we often do not know where discrepancies between the model's prediction and the actual outcome occur prior to observing executions in the real-world. One way to address this is to bias the planner away from these discrepancies by inflating the cost of states and actions where we previously observed the model to be inaccurate. Making such decisions about where and how to bias purely at the planning-level, however, neglects valuable information from the control-level, which gives a more fine-grained understanding of where and how the model went wrong during execution. Based on this observation, our key idea is to first infer a statistical model over discrepancies in the control-level's model. Then, we translate this model to the planning-level, where we use it to more informatively bias the planner away from states and actions where the model's predicted outcome is likely to be inaccurate. We demonstrate that our framework enables a robot to complete tasks, despite an inaccurate planning model, with greater efficiency than existing approaches. We do so through an experimental evaluation in simulation and real-robot experiments on NASA's Astrobee free-flyer.

Real-world autonomous systems often employ probabilistic predictive models of human behavior during planning to reason about their future motion. Since accurately modeling human behavior a priori is challenging, such models are often parameterized, enabling the robot to adapt predictions based on observations by maintaining a distribution over the model parameters. Although this enables data and priors to improve the human model, observation models are difficult to specify and priors may be incorrect, leading to erroneous state predictions that can degrade the safety of the robot motion plan. In this work, we seek to design a predictor which is more robust to misspecified models and priors, but can still leverage human behavioral data online to reduce conservatism in a safe way. To do this, we cast human motion prediction as a Hamilton-Jacobi reachability problem in the joint state space of the human and the belief over the model parameters. We construct a new continuous-time dynamical system, where the inputs are the observations of human behavior, and the dynamics include how the belief over the model parameters change. The results of this reachability computation enable us to both analyze the effect of incorrect priors on future predictions in continuous state and time, as well as to make predictions of the human state in the future. We compare our approach to the worst-case forward reachable set and a stochastic predictor which uses Bayesian inference and produces full future state distributions. Our comparisons in simulation and in hardware demonstrate how our framework can enable robust planning while not being overly conservative, even when the human model is inaccurate. Videos of our experiments can be found at the project website1.

Many problems in robotics involve multiple decision making agents. To operate efficiently in such settings, a robot must reason about the impact of its decisions on the behavior of other agents. Differential games offer an expressive theoretical framework for formulating these types of multi-agent problems. Unfortunately, most numerical solution techniques scale poorly with state dimension and are rarely used in real-time applications. For this reason, it is common to predict the future decisions of other agents and solve the resulting decoupled, i.e., single-agent, optimal control problem. This decoupling neglects the underlying interactive nature of the problem; however, efficient solution techniques do exist for broad classes of optimal control problems. We take inspiration from one such technique, the iterative linear-quadratic regulator (ILQR), which solves repeated approximations with linear dynamics and quadratic costs. Similarly, our proposed algorithm solves repeated linear-quadratic games. We experimentally benchmark our algorithm in several examples with a variety of initial conditions and show that the resulting strategies exhibit complex interactive behavior. Our results indicate that our algorithm converges reliably and runs in real-time. In a three-player, 14-state simulated intersection problem, our algorithm initially converges in <; 0.25 s. Receding horizon invocations converge in <; 50 ms in a hardware collision-avoidance test.

A significant challenge for the practical application of reinforcement learning to real world problems is the need to specify an oracle reward function that correctly defines a task. Inverse reinforcement learning (IRL) seeks to avoid this challenge by instead inferring a reward function from expert demonstrations. While appealing, it can be impractically expensive to collect datasets of demonstrations that cover the variation common in the real world (e.g. opening any type of door). Thus in practice, IRL must commonly be performed with only a limited set of demonstrations where it can be exceedingly difficult to unambiguously recover a reward function. In this work, we exploit the insight that demonstrations from other tasks can be used to constrain the set of possible reward functions by learning a "prior" that is specifically optimized for the ability to infer expressive reward functions from limited numbers of demonstrations. We demonstrate that our method can efficiently recover rewards from images for novel tasks and provide intuition as to how our approach is analogous to learning a prior.

Learning from demonstration (LfD) is a common technique applied to many problems in robotics, such as populating grasp databases, training for reinforcement learning of high-level skill sets and bootstrapping motion planners. While such approaches are generally highly valued, they rely on the often time-consuming process of gathering user demonstrations, and hence it becomes difficult to attain a sizeable dataset. In this paper, we present a tool capable of recording large numbers of high-dimensional demonstrations of mobile manipulation tasks provided by non-experts in the field. Our tool accomplishes this via a web interface that requires no additional software to be installed beyond a web browser, as well as a scalable architecture that is capable of supporting 10 concurrent demonstrators on a single server. Our architecture employs a lightweight simulation environment to reduce unnecessary computations and improve performance. Furthermore, we show how our tool can be used to gather a large set of demonstrations of a mobile manipulation task by leveraging existing crowdsource platforms. The data set collected has been made available to the robotics community. We also present experiments in which we apply demonstrations collected through our infrastructure to teach a robot how to grasp, to teach a robot how to perform dexterous manipulation tasks such as scooping and to accelerate motion planning for full-body manipulation tasks.