Humans are apt at navigating the world while avoiding bumping into objects and other humans along the way. While this ability comes effortlessly to humans, enabling robots to navigate their environment like humans do remains an active research area.
The challenges to robot navigation are multi-fold; while some stem from the noise in robots' sensors, others are due to robots' mechanical imperfections, and the difficulty in predicting the movements and intentions of humans and other moving objects in space.
To help curb some of these challenges, the AI Research Lab within Nokia Bell Labs is designing learning-based solutions to improve robots’ understanding of the world and enable them to take better navigation decisions.
Our work has led to the design of a state-of-the-art Reinforcement Learning solution that enables robots to gracefully navigate around static obstacles and tight pathways. Further, to enhance robots' ability to navigate around dynamic obstacles, we have trained models to predict the travel times of trail-following and free-space robots.
Our data-driven approach has been proven to enable robots to accurately predict future conflicts and take corrective measures to circumvent them.
A significant challenge in the robotics field, which is often overlooked in academic settings, is ensuring that the solutions work in the real world with real physical robots. Here, at Nokia Bell Labs, we are adamant about making our solutions work in practice. Key enablers for our work are the robotic environments at Nokia Bell Labs which consist of an elaborate ground and flying robots' playground at their Murray Hill campus, and the ground robot deployment at the Nokia Chennai Factory in India.
Enabling safe and autonomous robot navigation opens a realm of future applications that can boost the economy and shield humans from dangerous and strenuous working conditions.
