Touch-based Sensing for Evaluating Vegetation in Complex Navigation Environments

Project begins 2023.

Detecting, classifying, and overriding/avoiding vegetation is critical challenge in off-road autonomous navigation (Kelly, et al., 2006). Vegetation can affect autonomous navigation in two main ways. First, it can affect the AGV perception by obscuring potential obstacles or drivable regions. Second, it can affect the AGV mobility by resisting the vehicle motion. This second factor has been studied extensively by the US Army for single trees acting on the pushbar of a passenger vehicle on flat terrains (Mason, Gates, & Moore, 2012) and on slopes (Rybansky, 2020). More recently, (Wasfy, Wasfy, Paramsothy, & Sanikommu, 2020) have developed physics-based simulation of vegetation override by representing stems as thin beams in an FEA solver. While both the empirical and first-principles approaches have been fruitful, the fact remains that all previous methods drastically simplify the complexity of the true off-road vegetation environment.

A self-driving vehicle must identify and assess the risk of a complex mixture of vegetation. Ultimately, identifying this risk requires predicting the forces exerted by the vegetation on the vehicle. In some cases, vegetation should be avoided; in other cases, the vehicle can easily push through with no damage – this is known as vegetation override. However, it is currently far beyond the state-of-the-art to calculate vegetation forces using exteroceptive (camera or lidar) sensors for complex terrains. Furthermore, even the most cutting-edge physics-based approaches (Wasfy, Wasfy, Paramsothy, & Sanikommu, 2020) are a drastic simplification of the real-world problem.

In this work, we seek to overcome this gap in the current capabilities for off-road navigation by developing a “touch-based” sensor for measuring the forces exerted by the vegetation on the vehicle. The sensor will be exteroceptive in that it will be external to the vehicle system – and thus take measurements that are easily transferrable between vehicles – but proprioceptive in that it measures the forces on the vehicle through direct contact. In the proposed research, we will investigate how this new sensor modality can be used to improve and augment current approaches for predicting risk in complex vegetated environments.

The fundamental research questions of this work are a) Can the risk associated with traversing complex vegetation be accurately assessed using touch-sensing? and b) Can touch-based measurements be used to complement traditional sensing modalities (camera, lidar) to estimate the risk posed by navigation in vegetation? Said another way, can touch based sensors distinguish features of vegetation (e.g., stiffness, resistance over long pushes) that exteroceptive sensors cannot? Our objective is to measure vegetation override forces in a way that is independent of the vehicle. That is, we will not use tire slip, vehicle acceleration, or any other proxy for exerted force that involves the vehicle platform. Rather we will use direct measurements of the force, as measured by a pushbar with embedded force sensors, to directly measure the override forces.

References:

• Kelly, A., Stentz, A., Amidi, O., Bode, M., Bradley, D., Diaz-Calderaon, A., . . . Rander, P. (2006). Toward reliable off road autonomous vehicles operating in challenging environments. International Journal of Robotics Research, 25(5-6), 449-483.
• Mason, G. L., Gates, B. Q., & Moore, V. D. (2012). Determining forces required to override obstacles for ground vehicles. Journal of Terramechanics, 49(3-4), 191-196.
• Rybansky, M. (2020). Determination the ability of military vehicles to override vegetation. Journal of Terramechanics, 91, 129-138.
• Wasfy, T., Wasfy, H., Paramsothy, J., & Sanikommu, S. (2020). Finite Element Model for Prediction of Ground Vehicle Mobility Over Vegetation Covered Terrains. International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (p. V002T02A024). American Society of Mechanical Engnieers.

#1.40