RoboStack: Using the Robot Operating System alongside the Conda and Jupyter Data Science Ecosystems

We argue that it is beneficial to tightly couple the widely-used Robot Operating System with Conda, a cross-platform, language-agnostic package manager, and Jupyter, a web-based interactive computational environment affording scientific computing. We provide new ROS packages for Conda, enabling the installation of ROS alongside data-science and machine-learning packages with ease. Multiple ROS versions (currently Melodic, Noetic and ROS2 Foxy) can run simultaneously on one machine, with pre-compiled binaries available for Linux, Windows and OSX, and the ARM architecture (e.g. the Raspberry Pi and the new Apple Silicon). To deal with the large size of the ROS ecosystem, we significantly improved the speed of the Conda solver and build system by rewriting the crucial parts in C++. We further contribute a collection of JupyterLab extensions for ROS, including plugins for live plotting, debugging and robot control, as well as tight integration with Zethus, an RViz like visualization tool. Taken together, RoboStack combines the best of the data-science and robotics worlds to help researchers and developers to build custom solutions for their academic and industrial projects.

Multimodal Probabilistic Model-Based Planning for Human-Robot Interaction

This paper presents a method for constructing human-robot interaction policies in settings where multimodality, i.e., the possibility of multiple highly distinct futures, plays a critical role in decision making. We are motivated in this work by the example of traffic weaving, e.g., at highway on-ramps/off-ramps, where entering and exiting cars must swap lanes in a short distance---a challenging negotiation even for experienced drivers due to the inherent multimodal uncertainty of who will pass whom. Our approach is to learn multimodal probability distributions over future human actions from a dataset of human-human exemplars and perform real-time robot policy construction in the resulting environment model through massively parallel sampling of human responses to candidate robot action sequences. Direct learning of these distributions is made possible by recent advances in the theory of conditional variational autoencoders (CVAEs), whereby we learn action distributions simultaneously conditioned on the present interaction history, as well as candidate future robot actions in order to take into account response dynamics. We demonstrate the efficacy of this approach with a human-in-the-loop simulation of a traffic weaving scenario.