Challenges in Scene Understanding for Autonomous Systems

Autonomous robots and Machine Learning have played a pivotal role in solving many real-world problems over the past decade. In all applications of learning and exploration-based algorithms for robots, one has to incorporate scene understanding to enable the robot to be aware of the environment in which it is functioning. This imposes inherent importance on the development of accurate and efficient scene understanding algorithms. Although methods like semantic segmentation deliver commendable performance in most settings, some noticeable setbacks exist. In this work, we present various limitations and drawbacks faced by current autonomous pipelines along with solutions to mitigate the same.