IESA AI Summit 2021 Apr 07 - Session 1 - AIML Applications for Robotics - Vijay R Seelam

In this talk

1. 5:25

   Why factories need more adaptable robots

   Seelam says industrial robots are concentrated in fields such as automotive and semiconductor manufacturing, where work is highly structured and machines often operate behind safety barriers. Their cost, specialist programming needs, and sensitivity to disruptions limit adoption in smaller or less standardized operations. He argues that robots should complement people by handling repetitive tasks while workers focus on judgment and creative work.

2. 10:46

   Three kinds of intelligence for robots

   Seelam groups robotics challenges into ambient intelligence, or understanding the surroundings; contextual intelligence, or responding to changes; and task intelligence, or learning the details of a job. He says environmental awareness has advanced through machine vision, while adapting to changing circumstances and understanding task features such as shape, color, and grip remain difficult.

3. 13:35

   A four-part AI and robotics framework

   Seelam outlines sensing, perception, decision and policy, and execution as the stages that connect AI models to physical robot actions. Robots need to combine their own sensors with information from other equipment, then plan movements while accounting for safety rules and coordination with other robots. Edge and cloud computing each raise questions about processing capacity, system integration, and communications reliability.

4. 19:50

   Applications in handling, assembly, and inspection

   Seelam describes object detection and obstacle avoidance as ways to help robots work around people, other machines, and imperfectly positioned fixtures. For manipulation, robots must adapt grip and force to an object's shape and material; tactile feedback can help with delicate electronics and gear assembly. Robot-based inspection can find assembly defects during production, check surface quality, and verify functions such as a touchscreen.

5. 26:44

   Learning from people and task data

   Seelam expects robotics to become more common as hardware, AI, and computing improve, while noting constraints such as cost, heat, size, processing power, and reliable communication. He says training models is difficult when data is scarce, especially in less structured settings and smaller industries. His proposed approach has people teach robots through task examples, with both improving as more data is collected.