AI in Robotics and Reproducible Data Science on an Industrial Scale

# SCHEDULE #

18.00 Pawel Rosikiewicz, SwissAI
SwissAI in 2019

18.05 Nadia FIGUEROA, EPFL
Teaching robots complex manipulation tasks from demonstrations

18.40 Changan CHEN, EPFL
Crowd-aware Robot Navigation with Attention-based Deep Reinforcement Learning

19.10 Dushyant KHOSLA, PMI
Enterprise-ready ML: Lessons from the Frontline for doing Reproducible and Rapid Data Science

19.50 Apero

# ABSTRACTS #

Nadia FIGUEROA
Humans have a remarkable way of learning, adapting and mastering new tasks, such as manipulating objects or tools. With the current advances in Artificial Intelligence and Machine Learning (ML), the promise of having robots with such capabilities seems to be on the cusp of reality. Teaching such skills to robots, however, is complicated as they involve a level of complexity that cannot be tackled by classical ML methods in an unsupervised way. We specifically address the problem of learning complex (continuous) manipulation tasks from human demonstrations. Our proposed approaches are validated on a variety of tasks that involve single-target complex motions with a KUKA LWR 4+ robot arm.

Changan CHEN
Mobility in an effective and socially-compliant manner is an essential yet challenging task for robots operating in crowded spaces. Recent works have shown the power of deep reinforcement learning techniques to learn socially cooperative policies. However, their cooperation ability deteriorates as the crowd grows since they typically relax the problem as a one-way Human-Robot interaction problem. In this work, we want to go beyond first-order Human-Robot interaction and more explicitly model Crowd-Robot Interaction (CRI). We propose to (i) rethink pairwise interactions with a self-attention mechanism, and (ii) jointly model Human-Robot as well as Human-Human interactions in the deep reinforcement learning framework. Our model captures the Human-Human interactions occurring in dense crowds that indirectly affects the robot's anticipation capability. Our proposed attentive pooling mechanism learns the collective importance of neighboring humans with respect to their future states. Various experiments demonstrate that our model can anticipate human dynamics and navigate in crowds with time efficiency, outperforming state-of-the-art methods.

Dushyant KHOSLA
The explosion in the volume, variety, and velocity of data was quickly followed by a surge of interest in learning Data Science. Bootcamps, MOOCs and even university degrees were offered to train people in the art and craft of building ML/AI models for the industry. The number of professionals capable of doing data science rose, but the quality fell. Today in the industry we have teams of DevOps, Data Engineers, and Data Scientists caught in a cycle of confusion because there is often little understanding of what the other person does. Consequently, a lot of smart ML models never make it into production. One solution to this problem is to hire people with skills in all three domains, but they're notoriously difficult to find. The other is to make it easy for Data Scientists to build models that can be easily integrated into a software system. Through collaborative efforts between our data teams and its enablers, we engineered delivery process built with open-source tools that allowed us to move quickly from ideas to the product - all within the legal and privacy constraints of a large organization. Today, we are here to present the tools of our trade, to talk about the problems it helps us solve and to inspire you to adopt some of these principles and practices to create a flexible, inspectable and reproducible for building Data Products for the industry.

Sponsors:
EPFL, Innovaud and Phillip Morris International