Unsupervised Machine Learning Certifiable Object Detection for Autonomous Aircraft Detect and Avoid

Istari, partnering with the Sensing Perception Autonomy and Robot Kinetics (SPARK) Lab at MIT, proposes to study the feasibility of integrating pose estimation and object detection algorithms developed at MIT into the Air Force digital aircraft design certification procedure to target improved model performance.  The SPARK Lab has developed highly accurate methods to certify pose estimations and object detection for autonomous vehicles.   In particular, it was shown that rotation estimation can be decoupled from the estimation of the object translation and shape, and it was demonstrated that (i) the optimal object rotation can be computed via a tight (small-size) semidefinite relaxation, and (ii) the translation and shape parameters can be computed in closed-form given the rotation. An outlier rejection layer to the solver was also added, hence making it robust to a large number of misdetections. The optimal solver is then wrapped in a robust estimation scheme based on graduated non-convexity. To further enhance robustness to outliers, the first graph-theoretic formulation to prune outliers in category-level perception was developed, which removes outliers via convex hull and maximum clique computations; the resulting approach is robust to 70 − 90% outliers. The solver is then combined with a deep-learned keypoint detector, and the resulting approach improves over the state of the art in vehicle pose estimation.