Challenges in Human-Robot Interaction: "Robots as Agents", interview with Sunglok Choi

Social robots are soon going to drastically impact on our lives, changing the way we work and spend our leisure time at home. Though I discussed this theme often in my previous posts, I have until now not spoken of one very important --yet often ignored-- aspect: mobility and accessibility. It is impossible to serve humans or engage in any type of human-robot interaction if the robots are not near to humans or able to reach them, this is obvious! Even though pop-culture stereotypically shows robots that can autonomously go everywhere, the reality is very different and a simple task like walking or climbing stairs can be very challenging.

Mr. Choi, robotic research scientist at the Robot and Cognitive System Department of ETRI (one of the most famous Korean research institutions), is working with his team to enhance robot mobility. The challenge is how to make robots robustly and autonomously navigate thorough environments that are dynamic and unpredictable because they are meant for humans, as for example a street with cars or a room full of people. At ETRI researchers have worked on different strategies based on hardware and software technology in order to simplify the environmental randomness and allow robust navigation.

At ETRI, researchers built software to algorithmically stabilize and overcome the noise generated from the environment when using computer vision systems. They created software to compute robust regression and visual odometry (estimation of motion and location of a moving camera from optical flow) as well as a freely available navigation library named uRon. Moreover, they built a hardware optical system named StarLITE, which by tracking LED markers on the ceiling, can compute the robot indoor location (like an in-door GPS system); they also developed a system to generate accurate maps of indoor environments using a robot cart that can scan and reconstruct the shape of a room based on a laser scanning sensor. These technologies together are then seamlessly applied to different types of social robots meant to serve humans in public spaces (the Tomorrow City project) in a variety of tasks (e.g., guidance and advertising), as exemplified by the images here below.

The approach followed by ETRI is particularly interesting for one reason: not only do they attempt to reduce the complexity of the navigation task by reducing the complexity of the environment with ad-hoc hardware (StarLITE) but they particularly focus on the development of new algorithms to stabilize current navigation techniques. In Mr. Choi's opinion, in fact, in the next 20 years it is unlikely we will see many hardware breakthroughs, so we should rather focus our efforts on the improvement of our software. In fact, AI agents are a sort of robot, though completely virtual, and can be improved in terms of perception (recognizing and distinguishing objects), action (intention recognition and proactivity) and connectivity (crowd systems for robots).

In Mr. Choi's words: "In the next 20 years, we will see a radical improvement of algorithms for robots (especially in perception), dictated by algorithmic and maybe some hardware breakthroughs. These improvements will most likely impact many other devices (e.g. TV, mobile phone, automobile, and so on), transforming those devices as well into intelligent cognitive agents -or in other terms- turning them into a new type of robot."