Recent robotic manipulation competitions have highlighted that sophisticated robots still struggle to achieve fast and reliable perception of task-relevant objects in complex, realistic scenarios. To improve these systems' perceptive speed and robustness, we present SegICP, a novel integrated solution to object recognition and pose estimation. SegICP couples convolutional neural networks and multi-hypothesis point cloud registration to achieve both robust pixel-wise semantic segmentation as well as accurate and real-time 6-DOF pose estimation for relevant objects. Our architecture achieves 1 cm position error and <; 5° angle error in real time without an initial seed. We evaluate and benchmark SegICP against an annotated dataset generated by motion capture.

This paper presents a new method for injecting human inputs into mixed-initiative interactions between humans and robots. The method is based on a model-predictive control (MPC) formulation, which inevitably involves predicting the system (robot dynamics as well as human input) into the future. These predictions are complicated by the fact that the human is interacting with the robot, causing the prediction method itself to have an effect on future human inputs. We investigate and develop different prediction schemes, including fixed and variable horizon MPCs and human input estimators of different orders. Through a search-and-rescue-inspired human operator study, we arrive at the conclusion that the simplest prediction methods outperform the more complex ones, i.e., in this particular case, less is indeed more.

In this paper, we present a control theoretic formulation for composing human control inputs with an automatic controller for shared control of a quadruped rescue robot. The formulation utilizes a model predictive controller to guide human controlled leg positions to satisfy state constraints that correspond to static stability for the robot. A hybrid control architecture that incorporates the model predictive controller is developed to implement a gait that guarantees stable foot-placements for the robot. The algorithm is applied to a simulation of a quadruped rescue robot with human input provided through haptic joysticks.

This paper presents a control theoretic formulation and optimal control solution for integrating human control inputs subject to linear state constraints. The formulation utilizes a receding horizon optimal controller to update the control effort given the most recent state and human control input information. The novel solution to the corresponding finite horizon optimal control problem with terminal constraint is derived using Hilbert space methods. The control laws are applied to two planar human-driven mass-cart pendula, where the task is to synchronize the pendula's oscillations.