Aug 13 2020
A team of researchers, from the Army Research Laboratory of the U.S. Army Combat Capabilities Development Command and also from The University of Texas at Austin, has created a new algorithm that enables a self-driving ground vehicle to enhance its prevalent navigation systems by observing the way a human drives a vehicle.
Army researchers use human teachers to improve navigation in autonomous systems. Image Credit: U.S. Army photo.
The researchers tested their method - known as adaptive planner parameter learning from demonstration (APPLD) - on one of the experimental self-driving ground vehicles belonging to the Army.
Using approaches like APPLD, current Soldiers in existing training facilities will be able to contribute to improvements in autonomous systems simply by operating their vehicles as normal. Techniques like these will be an important contribution to the Army’s plans to design and field next-generation combat vehicles that are equipped to navigate autonomously in off-road deployment environments.
Dr Garrett Warnell, Researcher, Army Research Laboratory, U.S. Army Combat Capabilities Development Command
The team combined machine learning from demonstration algorithms and more traditional self-driving navigation systems. Instead of substituting a traditional system collectively, the APPLD system learns how to adjust the prevalent system to act almost like the human's demonstration.
Such an example enables the deployed system to retain all the advantages of traditional navigation systems such as safety, explainability, and optimality, and at the same time, enables the system to be adaptable and flexible to new settings, added Warnell.
A single demonstration of human driving, provided using an everyday Xbox wireless controller, allowed APPLD to learn how to tune the vehicle’s existing autonomous navigation system differently depending on the particular local environment.
Dr Garrett Warnell, Researcher, Army Research Laboratory, U.S. Army Combat Capabilities Development Command
Dr Warnell continued, “For example, when in a tight corridor, the human driver slowed down and drove carefully. After observing this behavior, the autonomous system learned to also reduce its maximum speed and increase its computation budget in similar environments. This ultimately allowed the vehicle to successfully navigate autonomously in other tight corridors where it had previously failed.”
The study is part of the Open Campus initiative of the Army. This initiative allows Texas-based Army researchers to work along with academic partners at The University of Texas at Austin.
APPLD is yet another example of a growing stream of research results that has been facilitated by the unique collaboration arrangement between UT Austin and the Army Research Lab. By having Dr Warnell embedded at UT Austin full-time, we are able to quickly identify and tackle research problems that are both cutting-edge scientific advances and also immediately relevant to the Army.
Dr Peter Stone, Professor and Chair of the Robotics Consortium, The University of Texas at Austin
The experiments conducted by the team demonstrated that, post-training, the APPLD system successfully navigated the test settings more rapidly and with fewer failures when compared to the traditional system. It was also observed that the trained APPLD system usually navigated the setting more quickly than the human who trained it.
The researchers’ study titled “APPLD: Adaptive Planner Parameter Learning From Demonstration” was published in IEEE Robotics and Automation Letters—a peer-reviewed journal.
“From a machine learning perspective, APPLD contrasts with so called end-to-end learning systems that attempt to learn the entire navigation system from scratch,” added Dr Stone.
He further stated, “These approaches tend to require a lot of data and may lead to behaviors that are neither safe nor robust. APPLD leverages the parts of the control system that have been carefully engineered, while focusing its machine learning effort on the parameter tuning process, which is often done based on a single person’s intuition.”
The APPLD system denotes a novel example, where individuals lacking expert-level know-how in robotics can help train and enhance self-driving vehicle navigation in a wide range of settings.
But instead of having small groups of engineers to physically adjust the navigation systems in an insignificant number of test settings, an almost infinite number of users would be able to offer the required data to the system, enabling it to adjust itself to an unlimited number of settings.
“Current autonomous navigation systems typically must be re-tuned by hand for each new deployment environment,” stated Dr Jonathan Fink, an Army researcher. “This process is extremely difficult—it must be done by someone with extensive training in robotics, and it requires a lot of trial and error until the right systems settings can be found.”
Dr Fink continued, “In contrast, APPLD tunes the system automatically by watching a human drive the system—something that anyone can do if they have experience with a video game controller. During deployment, APPLD also allows the system to re-tune itself in real-time as the environment changes.”
To improve the Next Generation Combat Vehicle, the Army will design robotic combat vehicles as well as optionally manned fighting vehicles that can navigate independently even in off-road deployment settings.
Although soldiers can drive present-day combat vehicles to navigate such environments, the settings continue to be extremely challenging for advanced self-driving navigation systems. The APPLD and analogous methods offer a new promising way for the Army to enhance the capabilities of present-day autonomous navigation.
“In addition to the immediate relevance to the Army, APPLD also creates the opportunity to bridge the gap between traditional engineering approaches and emerging machine learning techniques, to create robust, adaptive, and versatile mobile robots in the real-world,” stated Dr Xuesu Xiao, the study’s lead author and a postdoctoral researcher at The University of Texas at Austin.
To sustain this study, the researchers will test the APPLD system in a wide range of outdoor settings, make use of soldier drivers, and try out a broader range of prevalent autonomous navigation methods.
The team will also investigate whether the inclusion of more sensor data such as camera images can result in learning more intricate behaviors, for example, adjusting the navigation system to work under different kinds of conditions, such as with other objects present or on different lands.
Journal Reference:
Xiao, X., et al. (2020) APPLD: Adaptive Planner Parameter Learning From Demonstration. IEEE Robotics and Automation Letters. doi.org/10.1109/LRA.2020.3002217.