Smart software replaces expensive sensors for glass wall detection with 96% accuracy

A analysis group has developed autonomous driving software program that enables cheap sensors to detect clear obstacles equivalent to glass partitions, offering a substitute for high-performance sensors. This expertise can be utilized in present robots, negating the necessity for added gear whereas guaranteeing detection efficiency that is the same as that supplied by costly typical gear.

The paper is printed within the journal IEEE Transactions on Instrumentation and Measurement. The group was led by Professor Kyungjoon Park on the Division of Electrical Engineering and Laptop Science, Daegu Gyeongbuk Institute of Science & Expertise.

Autonomous driving robots usually use LiDAR sensors to detect their environment and navigate. Functioning as “laser eyes,” costly LiDAR sensors decide distance and construction by projecting gentle and measuring reflection time.

Cheap LiDAR sensors can not detect clear objects equivalent to these manufactured from glass; they might mistake them for empty house, probably leading to a collision. Excessive-resolution ultrasonic LiDAR sensors or cameras wouldn’t have this limitation, however their use will increase system complexity and raises prices by tons of of hundreds to tens of millions of gained.

To offer an alternate, a DGIST analysis group led by Professor Kyungjoon Park developed probabilistic incremental navigation-based mapping (PINMAP), an algorithm that approaches problem-solving by way of software program, not {hardware}. PINMAP accumulates uncommon level information that cheap LiDAR sensors can detect solely sporadically. Utilizing these information, PINMAP probabilistically calculates the chance of the presence of glass partitions over time.

The PINMAP algorithm relies on Cartographer (map charting) and Nav2 (navigation), that are open-source instruments which might be broadly used within the ROS 2 ecosystem. PINMAP has the benefit of straightforward applicability whereas eliminating the necessity to change the present system construction.

As an alternative of upgrading the sensors at a excessive value, the algorithm alters the best way the present sensors deal with information; that’s, it makes use of software program to enhance the detection efficiency of cheap LiDAR sensors.

In a real-world experiment carried out at DGIST, PINMAP detected glass partitions with 96.77% accuracy, which is properly above the practically 0% detection charge of the normal strategy utilizing the identical cheap LiDAR sensors (Cartographer-SLAM). The software program distinction that PINMAP presents demonstrated an incredible efficiency increase.

Professor Park stated, “PINMAP flips the standard knowledge that {hardware} efficiency equals system efficiency and proposes a brand new customary whereby software program can enhance sensor capabilities. This examine reveals that guaranteeing steady autonomous driving is feasible with out counting on high-performance gear.”

The algorithm the analysis group developed presents a considerable financial benefit as a result of it achieves detection efficiency corresponding to that of pricey LiDAR sensors at lower than one-tenth of the associated fee. This expertise is predicted to scale back collisions between autonomous driving robots and glass or clear acrylic partitions in indoor areas equivalent to hospitals, airports, procuring malls, and warehouses, thus contributing to the large-scale deployment of service robots.