Dynamic path planning for reconfigurable rovers using a multi-layered grid

https://doi.org/10.1016/j.engappai.2019.08.011Get rights and content

Abstract

Autonomy on rovers is desirable in order to extend the traversed distance, and therefore, maximize the number of places visited during the mission. It depends heavily on the information that is available for the traversed surface on other planet. This information may come from the vehicle’s sensors as well as from orbital images. Besides, future exploration missions may consider the use of reconfigurable rovers, which are able to execute multiple locomotion modes to better adapt to different terrains. With these considerations, a path planning algorithm based on the Fast Marching Method is proposed. Environment information coming from different sources is used on a grid formed by two layers. First, the Global Layer with a low resolution, but high extension is used to compute the overall path connecting the rover and the desired goal, using a cost function that takes advantage of the rover locomotion modes. Second, the Local Layer with higher resolution but limited distance is used where the path is dynamically repaired because of obstacle presence. Finally, we show simulation and field test results based on several reconfigurable and non-reconfigurable rover prototypes and a experimental terrain.

Introduction

Autonomy is an essential capability for rovers to explore the surface of other planets. The distances from Earth entail big latencies in communications between the rover and the terrestrial ground station. As an example, there is a radio transmission delay of several minutes between Earth and Mars (Lester and Thronson, 2011, Lester et al., 2017). Therefore, direct teleoperation arises as a difficult task to be carried out remotely from Earth. Besides, communications with rovers at the red planet generally occur only a few times per Martian sol (solar day) due to the availability of Deep Space Network antennas, conforming a limited time-slot for providing commands and retrieving data from the rover (Bajracharya et al., 2008). These facts are contrary to the necessity of increasing the number of scientifically interesting places visited by rovers. Providing higher autonomy would allow them to traverse longer distances. However, new issues arise since rovers tackle a high uncertainty when they are traveling, i.e., they may encounter unexpected situations, mostly due to terrain shape and/or composition, as well as the existence of stones. These issues affect the traversability for the vehicle. The improper evaluation of the terrain could lead to a fatal situation of the vehicle, compromising the mission as a result. This was the case of the Spirit rover, which got stuck in loose sand, making it impossible to continue driving (Ono et al., 2015) and thus bringing the mission to an end. By using traversability information, autonomy can be improved thanks to the use of path planning algorithms, which allow the vehicle to compute onboard a safe path from one location to another.

Path planning has been used in Mars exploration missions along with rovers Spirit, Opportunity and Curiosity. A path planning approach, based on two levels (global and local), was deployed on these rovers (Maimone et al., 2007). The main reason behind it is to use data relative to rover surroundings while also considering information relative to the location of elements, such as obstacles, in other areas. As global planning algorithm, the Field-D* was used, initially introduced by Ferguson and Stentz (2006b). With this algorithm, a potential field is computed on a regular grid starting from the goal position to the rover location. Main particularity of this method is the use of an interpolation technique that assigns the values of this potential field to each node, based on the values of its already visited neighbors. By considering that edges connecting nodes are crossable by the path, it results to be smoother than previous methods like D*, where paths are strictly restricted to go through node locations. This algorithm has also re-planning capability, meaning the path can be updated during rover traverse in case the cost of any node is modified. This can happen any time the rover detects an obstacle on its way using its onboard sensors. Besides, this algorithm has been adapted to the use of multi-resolution grids (Ferguson and Stentz, 2006a) to minimize computational resources, saving computation in areas where the level of detail can be simplified. However, it is not clear neither the computational cost of this algorithm nor the steps to extract a path from it, leaving this operation to another algorithm that only focuses on the local planning. In the case of Mars rovers, the GESTALT local planner has been used (Carsten et al., 2007). It basically generates a series of arcs starting from the vehicle position, which are later evaluated according to the potential field created by the global planner. As result, the arc with the best evaluation is chosen to be followed by the rover.

An alternative path planning solution to Field-D* is the Fast Marching Method (FMM), originally introduced by Sethian (1999). It is a numerical method that solves the so-called eikonal equation, which is an expression defining the behavior of a wave that propagates over a continuous 2D scalar function — also extendable to 3D. Unlike Field-D*, only a quadratic expression can be used along each node of the grid to compute a continuous potential field. Such a potential field represents the arrival time of the wave at each location of the grid. Besides, in contrast to Field-D*, the computational cost of FMM is clearly stated, this being O(ζlogζ), where ζ is the number of nodes the grid is composed of. After computing the potential field, the path is extracted by just making use of a gradient descent method on it (Kimmel and Sethian, 2001, Liu and Bucknall, 2015). However, this path planning solution is not originally meant to include a re-planning capability. Previous research (Philippsen et al., 2008) has focused on the modification of FMM to make it dynamic, obtaining as result an algorithm called E*. Nevertheless, it does not consider the use of maps with multiple resolutions. An example of FMM using a multi-resolution map, but not dynamic re-planning, can be seen in the work of Petres et al. (2005), where an already-known environment is modeled with different levels of detail, obtaining as result sub-optimal but fast-computed paths.

In most cases, path planning algorithms consider traversability data to decide whether a path should go through a certain area or not. It heavily depends on the underlying physics of the terrain–vehicle interaction, i.e. terramechanics. Shape and composition of terrains determine the dynamic behavior of any body in contact with it (e.g. friction and slipping effects). The kinematic configuration of the vehicle, together with the distribution of its masses and inertias, affects the forces exerted on the surface, influencing on traction, as well as on the energy required to move the vehicle. While the terrain features cannot be changed, the rover kinematic configuration could be adapted to it. This is the case for vehicles categorized as reconfigurable, which are able to perform several locomotion modes, each one adapted to a particular terrain. A notable example of this feature is the ExoMars rover, which is being conceived to search for signs of life on Mars in the future ExoMars 2020 mission, lead by the European Space Agency (Vago et al., 2015, Vago et al., 2017). The particularity, with respect to other rovers, is the use of additional joints on top of its legs. Such joints are initially meant to deploy the wheels once the rover has landed on Mars, but later, they can be further used to improve traction on loose soil by the use of a locomotion mode called wheel-walking (Woods et al., 2009, Patel et al., 2010). In this sense, Azkarate et al. (2015) performed some experiments using an ExoMars rover prototype that demonstrated this statement. Other authors analyzed the performance of a similar locomotion mode, called push–pull (Creager et al., 2015), on loose soil as well, getting to the same conclusions. These modes of locomotion may be helpful in situations where otherwise, being only capable of executing the standard roving or normal driving locomotion mode, would result in the rover raising its power consumption or even getting entrapped. Having a good knowledge of the locomotion-terrain relation for a particular rover, a path planning algorithm could take advantage of this information, finding even more optimal paths. However, the design of an algorithm that takes into consideration different locomotion modes is still being investigated. In previous works, path planning algorithms have been developed aimed at the reconfiguration of the vehicle chassis. For example, Brunner et al., 2012, Brunner et al., 2014 proposed algorithms to find optimal paths adapting the chassis to overcome obstacles in the form of stairs, while Miró et al. (2010) carried out a research focused on maintaining the stability of the vehicle using FMM. However, none of these works takes into consideration a reconfigurable rover for long-range operations, such as those found in planetary exploration missions.

This paper proposes a path planning algorithm based on FMM that works with a multiple resolution grid made up of two layers. The role of these layers is fully explained in Section 2. The first one is used to compute the overall path using a cost function based on the locomotion performance according to the terrain features. The objective is to find the path that minimizes the power consumption, taking advantage of the use of multiple locomotion modes. The second layer is used to perform a novel repairing process that dynamically modifies the path, with the aim of avoiding any obstacle detected by the rover on the spot. Later on, simulation results are provided in Section 3 followed by results from a field test. They serve as a way to analyze the main particularities of the proposed algorithm within several planetary exploration situations. From these results, in Section 4 are extracted a series of conclusions and some ideas for future work.

Section snippets

Path planning

The proposed path planning algorithm is described in this section. Fig. 1 shows a schematic containing its different parts.

The first one is the Multi-layered Grid, which contains two layers, each of them with different resolution and range. First, the Global Layer uses information provided by satellites to determine the distribution of the different types of terrain present in the mission area. Orbital images, such as the Digital Elevation Maps (DEMs) from HiRISE (McEwen et al., 2007) or the

Results

Once the proposed path planning algorithm has been detailed, it is validated by means of simulations carried out using MATLAB software and a field test. The code used for the simulations and the real experiment can be found on GitHub repositories.1 ,2 The purpose of the first simulation test is to analyze the validity of the Global Path Planning on long-range navigation. In particular, data related

Conclusions

In this paper a novel path planning has been presented. It works at two scales, global and local, using the so called Multi-layered Grid. The grid is composed of two layers, each of them for different planning purposes. The path connecting the rover initial and goal locations is computed on the Global Layer, according to the information related to the terrain and the rover locomotion modes. Then, the generated global path plan is dynamically repaired using the Local Layer whenever an obstacle

Acknowledgments

This work was partially supported by the European Space Agency under contract number 4000118072/16/NL/LvH/gp and by the Spanish project DPI2015-65186-R.

References (31)

  • CunninghamC. et al.

    Improving slip prediction on mars using thermal inertia measurements

    Auton. Robots

    (2019)
  • FergasonR.L. et al.

    High-resolution thermal inertia derived from the thermal emission imaging system (THEMIS): thermal model and applications

    Journal of Geophysical Research: Planets

    (2006)
  • FergusonD. et al.

    Multi-resolution field D*

  • FergusonD. et al.

    Using interpolation to improve path planning: The Field D* algorithm

    J. Field Robotics

    (2006)
  • FilipJ. et al.

    Trajectory control for autonomous planetary rovers

  • Cited by (22)

    • Multi-obstacle path planning and optimization for mobile robot

      2021, Expert Systems with Applications
      Citation Excerpt :

      Saeed, Recupero, and Remagnino (2019) proposes a new method to solve the path planning problem in a static environment. On this basis, Ricardo Sanchez-Ibanez, Perez-Del-Pulgar, Azkarate, Gerdes, and Garcia-Cerezo (2019) proposes a path planning algorithm based on the fast-moving method. Environmental information from different sources is used in a two-tier grid.

    View all citing articles on Scopus

    No author associated with this paper has disclosed any potential or pertinent conflicts which may be perceived to have impending conflict with this work. For full disclosure statements refer to https://doi.org/10.1016/j.engappai.2019.08.011.

    View full text