William Vega-Brown

CELLO-EM: Adaptive sensor models without ground truth

We present an algorithm for providing a dynamic model of sensor measurements. Rather than depending on a model of the vehicle state and environment to capture the distribution of possible sensor measurements, we provide an approximation that allows the sensor model to depend on the measurement itself. Building on previous work, we show how the sensor model predictor can be learned from data without access to ground truth labels of the vehicle state or true underlying distribution, and we show our approach to be a generalization of non-parametric kernel regressors. Our algorithm is demonstrated in simulation and on real world data for both laser-based scan matching odometry and RGB-D camera odometry in an unknown map. The performance of our algorithm is shown to quantitatively improve estimation, both in terms of consistency and absolute accuracy, relative to other algorithms and to fixed covariance models.

Bayesian Learning for Safe High-Speed Navigation in Unknown Environments

In this work, we develop a planner for high-speed navigation in unknown environments, for example reaching a goal in an unknown building in minimum time, or flying as fast as possible through a forest. This planning task is challenging because the distribution over possible maps, which is needed to estimate the feasibility and cost of trajectories, is unknown and extremely hard to model for real-world environments. At the same time, the worst-case assumptions that a receding-horizon planner might make about the unknown regions of the map may be overly conservative, and may limit performance. Therefore, robots must make accurate predictions about what will happen beyond the map frontiers to navigate as fast as possible. To reason about uncertainty in the map, we model this problem as a POMDP and discuss why it is so difficult given that we have no accurate probability distribution over real-world environments. We then present a novel method of predicting collision probabilities based on training data, which compensates for the missing environment distribution and provides an approximate solution to the POMDP. Extending our previous work, the principal result of this paper is that by using a Bayesian non-parametric learning algorithm that encodes formal safety constraints as a prior over collision probabilities, our planner seamlessly reverts to safe behavior when it encounters a novel environment for which it has no relevant training data. This strategy generalizes our method across all environment types, including those for which we have training data as well as those for which we do not. In familiar environment types with dense training data, we show an 80% speed improvement compared to a planner that is constrained to guarantee safety. In experiments, our planner has reached over 8 m/s in unknown cluttered indoor spaces. Video of our experimental demonstration is available at http://groups.csail.mit.edu/rrg/bayesian_learning_high_speed_nav.

CELLO: A fast algorithm for Covariance Estimation

We present CELLO (Covariance Estimation and Learning through Likelihood Optimization), an algorithm for predicting the covariances of measurements based on any available informative features. This algorithm is intended to improve the accuracy and reliability of on-line state estimation by providing a principled way to extend the conventional fixed-covariance Gaussian measurement model. We show that in experiments, CELLO learns to predict measurement covariances that agree with empirical covariances obtained by manually annotating sensor regimes. We also show that using the learned covariances during filtering provides substantial quantitative improvement to the overall state estimate.

PROBE-GK: Predictive robust estimation using generalized kernels

Many algorithms in computer vision and robotics make strong assumptions about uncertainty, and rely on the validity of these assumptions to produce accurate and consistent state estimates. In practice, dynamic environments may degrade sensor performance in predictable ways that cannot be captured with static uncertainty parameters. In this paper, we employ fast nonparametric Bayesian inference techniques to more accurately model sensor uncertainty. By setting a prior on observation uncertainty, we derive a predictive robust estimator, and show how our model can be learned from sample images, both with and without knowledge of the motion used to generate the data. We validate our approach through Monte Carlo simulations, and report significant improvements in localization accuracy relative to a fixed noise model in several settings, including on synthetic data, the KITTI dataset, and our own experimental platform.

Admissible abstractions for near-optimal task and motion planning

We define an admissibility condition for abstractions expressed using angelic semantics and show that these conditions allow us to accelerate planning while preserving the ability to find the optimal motion plan. We then derive admissible abstractions for two motion planning domains with continuous state. We extract upper and lower bounds on the cost of concrete motion plans using local metric and topological properties of the problem domain. These bounds guide the search for a plan while maintaining performance guarantees. We show that abstraction can dramatically reduce the complexity of search relative to a direct motion planner. Using our abstractions, we find near-optimal motion plans in planning problems involving

Deep Inference for Covariance Estimation: Learning Gaussian Noise Models for State Estimation

10^{13}

Nonparametric Bayesian inference on multivariate exponential families

states without using a separate task planner.

Markov Chain Hallway and Poisson Forest Environment Generating Distributions

We present a novel method of measurement covariance estimation that models measurement uncertainty as a function of the measurement itself. Existing work in predictive sensor modeling outperforms conventional fixed models, but requires domain knowledge of the sensors that heavily influences the accuracy and the computational cost of the models. In this work, we introduce Deep Inference for Covariance Estimation (DICE), which utilizes a deep neural network to predict the covariance of a sensor measurement from raw sensor data. We show that given pairs of raw sensor measurement and ground-truth measurement error, we can learn a representation of the measurement model via supervised regression on the prediction performance of the model, eliminating the need for hand-coded features and parametric forms. Our approach is sensor-agnostic, and we demonstrate improved covariance prediction on both simulated and real data.