Home Bots & Brains Research: Risk Assessment in Autonomous Robots

Research: Risk Assessment in Autonomous Robots

by Pieter Werner

Researchers led by mechanical engineering professor Nader Motee at Lehigh University are addressing the challenges of ambiguity in robot perception to enhance autonomous decision-making. The project, funded by a $680,000 grant from the Office of Naval Research (ONR), aims to develop an innovative multi-stage, perception-based control paradigm that promises to make autonomous systems both safer and more efficient.

The endeavor is driven by the understanding that for robots to achieve full autonomy, they must be capable of conducting risk analysis similar to human decision-making processes. From deciding the speed of a vehicle based on road conditions to moderating tone in communication, humans constantly assess risks to make informed decisions. Robots, however, currently lack this nuanced ability.

Professor Motee, part of Lehigh University’s P.C. Rossin College of Engineering and Applied Science, emphasizes the critical need for robots to quantify the ambiguity in perception before conducting risk analysis. He explains that while human perception draws from past experiences, robots are limited by the number of samples they can process, leading to ambiguity and uncertainty. For instance, environmental factors such as rain, fog, or darkness can create uncertainty about an object’s identity, making it challenging for a robot to differentiate between a stop sign and a speed limit sign.

To address this, Motee and his team are focused on quantifying sources of uncertainty within machine learning models that use visual sensing. They aim to delve into the “black box” of various perception modules to comprehend how these models interpret their surroundings. By analyzing the relationship between input images and output labels, the researchers hope to quantify ambiguity and subsequently use risk measures for decision-making.

The end goal is to enable robots to assess risks accurately, thereby making informed decisions that contribute to effective communication and assistance in tasks such as disaster recovery. By analyzing and assessing risk, robots could potentially perceive human actions and determine how best to assist without causing hindrance.

Professor Motee is optimistic about the future, envisioning perception modules working cohesively as a network to create a smart society that can enhance health and lifestyle. By collaborating, these modules could contribute significantly to sectors like health, transportation, and security.

The research spearheaded by Professor Motee and his team represents a significant stride towards creating a world where robots can provide meaningful assistance while ensuring safety and efficiency.

Photo credit: Douglas Benedict/Academic Image

Misschien vind je deze berichten ook interessant