Perception system performance can vary dramatically with contextual factors such as environmental geometry, appearance, and other phenomena. In this work we present a theoretical framework for understanding the role of context in perception and discuss three approaches for predicting probabilistic performance from observations by efficiently learning local performance models. We compare these approaches with experiments on the monocular and stereo visual odometry systems for a ground robot, and show that they can effectively predict system failures in a wide variety of environments.

The popularity of cameras for perception is enabled in part by powerful intrinsic calibration routines, commonly requiring a user to manually collect images of a known calibration target. The manual nature of this process produces training data that is unevenly spread in the camera relative pose space. If we desire camera parameters that perform well on average over the entire relative pose space, training on such a dataset results in poor performance. To address this, we show that reasoning about the training data distribution to select a more uniformly-spread subset of images produces more accurate and stable calibrations with fewer images. Our approach can be used easily with most camera calibration algorithms. We demonstrate in large-scale physical experiments the effect of non-uniform training data and show that our approach outperforms baselines in reprojection error and parameter variance.

Assembly of large structures requires large fixtures, often referred to as monuments. Their cost and massive size limit flexibility and scalability of the manufacturing process. Numerous small mobile robots can replace these large structures and, therefore, replicate the efficiency of the assembly line with far more flexibility. An assembly line made up of mobile manipulators can easily and rapidly be reconfigured to support scalability and a varied product mix, while allowing for near optimal resource assignment. The challenge to using small robots in place of monuments is making their joint behavior precise enough to accomplish the task and efficient enough to execute subtasks in a reasonable period of time. In this paper, we describe a set of techniques that we combine to achieve the necessary precision and overall efficiency to build a large structure. We describe and demonstrate these techniques in the context of a testbed we implemented for assembling a wing ladder.

The ubiquitous additive Gaussian noise model is favored in statistical modeling applications for its flexibility and ease of use. Often noise is assumed to be well-represented by a constant covariance, while in reality error characteristics may change predictably. We present an efficient parametric covariance predictor based on the modified Cholesky decomposition that maps from features of the input to covariance matrices. In addition, we discuss fitting the predictor parameters using noise samples with simple regularization techniques. We demonstrate our approach by estimating observation covariances for range-bearing localization with simulated and experimental datasets and show that this results in increased filtering performance compared to traditional covariance adaptation and constant covariance baselines.

In the future, robotic surgical assistants may assist surgeons by performing specific subtasks such as retraction and suturing to reduce surgeon tedium and reduce the duration of some operations. We propose an apprenticeship learning approach that has potential to allow robotic surgical assistants to autonomously execute specific trajectories with superhuman performance in terms of speed and smoothness. In the first step, we record a set of trajectories using human-guided backdriven motions of the robot. These are then analyzed to extract a smooth reference trajectory, which we execute at gradually increasing speeds using a variant of iterative learning control. We evaluate this approach on two representative tasks using the Berkeley Surgical Robots: a figure eight trajectory and a two handed knot-tie, a tedious suturing sub-task required in many surgical procedures. Results suggest that the approach enables (i) rapid learning of trajectories, (ii) smoother trajectories than the human-guided trajectories, and (iii) trajectories that are 7 to 10 times faster than the best human-guided trajectories.