Object Perception

The article What Can Tesla TSLA And Waymo Learn From A Humans Ability To Focus, discusses some of the important differences between human perception systems and those employed by modern Autonomous Vehicles (AV). Recently, MIT and Toyota took a step towards human-like perception with the introduction of the DriveSeg research dataset.

What is DriveSeg ?

DriveSig has two parts. First, the data consists of 2 minutes and 47 seconds of high-resolution video captured during a daytime trip around the busy streets of Cambridge, Massachusetts. The videos 5,000 frames are densely annotated manually with per-pixel human labels of 12 classes of road objects. Second, a set of videos captured from a range of scenes drawn from MIT Advanced Vehicle Technology (AVT) Consortium data coarsely annotated through a novel semiautomatic annotation approach developed by MIT.

Ok, why is this interesting or important ?

Humans use motion as a very strong part of object recognition. For example, it is not unusual to recognize a friend from afar simply by recognizing their gait. Thus, motion is an important part of human object recognition, but to date has not been a big part of Autonomous Vehicle(AV) object recognition. Rather, AV object recognition has traditionally focused on training AI engines on individual images. With this dataset, researchers from Massachusetts Institute for Technology (MIT) AgeLab at the MIT Center for Transportation & Logistics and the Toyota Collaborative Safety Research Center (CSRC) want to enable AI research focused on using movement.

In sharing this dataset, we hope to encourage researchers, the industry, and other innovators to develop new insight and direction into temporal AI modeling that enables the next generation of assisted driving and automotive safety technologies, says Bryan Reimer, Research Scientist at the Agelab. Our long-standing working relationship with Toyota CSRC has enabled our research efforts to impact future safety technologies.

MIT has a significant capability in artificial intelligence research with its Computer Science and Artificial Intelligence Laboratory (CSAIL) laboratory. This lab has been active in the area of robotics and AI with fundamental work announcements such as the recent announcement of Giving soft robots feeling. In contrast, the Age Lab at MITs Center for Transportation emphasizes a multidisciplinary approach which they believe will be necessary to crack the reality of Autonomous Vehicles.

Engineers are investing significant creative and financial resources in developing higher levels of automation to support drivers to be ideally both increasingly comfortable and safe. However, true success requires a human-centered engineering perspective. Its more important than ever to leverage the expertise of other disciplines that amplify each others capabilities. For instance, computer scientists have the ability to mine and model large scale complex systems, while human factors engineers and psychologists are trained in developing an understanding of human behavior to ensure systems are optimally used in a way that enhances accessible and safe mobility, said Bryan Reimer.

To enable this effort, data was drawn from the Massachusetts Institute of Technologys Advanced Vehicle Technology (AVT) Consortium,an academic-industry partnership drawing together stakeholders across the automotive ecosystem to collaboratively invest in the development of real-world data to better understanding how drivers adapt to, use (or do not use), and behave with production level vehicle technologies including advanced driver assistance systems and automated driving. By encompassing data from a range of real-world driving scenarios, the DriveSeg dataset aims to provide variations in situations that facilitate deeper exploration of the value of motion in AV object recognition.

Overall, humans use concepts such as focus, abstraction, differential perception, and sensitivity to relative acceleration as key mechanisms to drive perception. Todays AI engines have tended to rely on brute-force methods with limited success. This MIT/Toyota CSRC combination should allow the use of an important additional learning dimension (movement) to advance the state-of-art forward.

Related Articles:

For those who would like more detailed technical papers, please visit my SAE work.

See the original post:
MIT And Toyota Announce An Important Dataset For Improving Perception Research - Forbes