QuijoteExpress - A novel APSI planning system for future space robotic missions

Communications delay, an undeterministic/dynamic environment, science return or cost efficiency are some of the reasons that claim for more autonomy on Space Missions. The main effort of previous Mars Rover missions (i.e., MER) have been focused on-board, where at least three systems have been successfully deployed: Autonomous navigation, SPOTTER (dust-devil & cloud detection) and AEGIS (autonomous data collection). Considering the sophistication of future rover payloads (e.g., Exomars Pasteur & Humboldt) and with the lessons learnt from previous rover missions, an advanced planner/replanner represents one of the main building blocks for enabling technology. This paper discusses a possible advanced planner candidate: QuijoteExpress an heuristic-driven planner based on a evolution of the APSI framework, APSI*. APSI is an ESA owned software framework to develop automated planning and scheduling applications. APSI* allows the definition of complex goals using a planning technique called Hierarchical Task Networks (HTN). The human operator can define in this way a plan in terms of complex goals while the planner is in charge of decomposing them into commands. This technique provides several advantages. First, it represents an improvement on the planner performance, as HTN simplifies the search space. The planner does not need to search anymore how to achieve repetitive tasks. Instead, predefined methods specify the different ways to accomplish complex goals. Second, it also simplifies the modeling which is one of the major problems that engineers need to face to deploy automated tools. Finally, it makes plans easier to understand and validate by humans. QuijoteExpress extends the AP2 planner developed in the frame of the ESA Goal Oriented Autonomous Controller Study (GOAC). It is divided in a strategic planner that leads the search and four tactical planners used to fix the plan. In QuijoteExpress, the problem can be evolved to different layers, being each layer more detailed than the previous one. It tries to extract a solution using an iterative repair technique as follows: First, it repairs the present layer by adding / deleting or moving elements of the plan. Once the layer is valid, the planner decomposes the problem to the next level and repeat the process. Besides the use of HTN, QuijoteExpress present other novelties. By analysing the structure of the problem, it can identify independent parts that will be then processed in parallel. This technique aims to improve the performance and might enable other techniques such as distributed computing. To manage the inherent uncertainty of rover missions, QuijoteExpress replaces the concept of valid plan by sufficient plan. It allows the user to represent problems for which some information is missing and the planner to find solutions for such problems. Finally, we have implemented a hierarchical model of a rover to test the new platform