34 th International Symposium on Automation and Robotics in Construction (ISARC 2017) Cartesian-Space Motion Planning for Autonomous Construction Machines M. Ragaglia a , A. Argiolas a and M. Niccolini a a Yanmar Research & Development Europe, Viale Galileo 3/A, 50125, Firenze, Italy E-mail: matteo_ragaglia@yanmar.com, alfredo_argiolas@yanmar.com, marta_niccolini@yanmar.com Abstract - Nowadays, the construction industry is probably the least productive and most dangerous among the various industry sectors. Given this scenario, it is quite clear that the introduc- tion of Autonomous Construction Machines (ACMs) could represent a great opportunity to improve both productivity and safety. To this purpose, a fundamental problem that has to be tackled is trajectory planning. In the last 15 years, sev- eral sample-based algorithms have been proposed, that relies on Joint-Space sampling. Unfortunately, this feature often results in trajectories that are quite counterintuitive from the point of view of a human being. In this work we propose “cart-RRT”, a Cartesian-Space randomized algorithm that improves the intuitiveness of the output trajectory, while en- suring both its safety (in terms of collision avoidance) and its feasibility. Keywords - Robotics; Motion Planning; Construction Machines; 1 Introduction Nowadays, in the construction industry, several opera- tions that require both high power and high accuracy (such as panel positioning, plumbing, material handling) are still manually performed by human workers, in very inefficient and dangerous ways. After a thorough field investigation, we believe that the construction industry could definitely benefit from the introduction of Autonomous Construction Machines (ACMs), in terms of increased productivity and human workers’ safety. Among the huge number of func- tionalities that are required to develop an ACM, motion planning plays a fundamental role. As a matter of fact, the planner has the responsibility to compute a trajectory that allows the ACM to travel from its original configuration to a desired one, while satisfying kino-dynamic constraints and avoiding collision with obstacles. Given the fact that robots are complex systems, of- ten characterized by nonlinear dynamics and actuation constraints, the problem of trajectory planning can eas- ily become computationally intractable. For this reason, randomized sample-based planning algorithms [7] have emerged as a reliable and appealing alternative to search- based planning techniques [9] and model predictive con- trol approaches [14]. Randomized sample-based planners like Rapidly-exploring Random Trees (RRT) [8] have ex- perienced a growing popularity since their introduction in the late ’90s. The main reason behind this success relies in a rather simple, yet effective idea: sample multiple robot configurations that do not entail collisions with obstacles and connect them to build either a graph or a tree of feasible trajectories. Then, a solution is extracted from this tree (or graph), in terms of a sequence of edges that connect cou- ples of nodes. Furthermore, it has been demonstrated in [1] that sample-based planning algorithms are probabilis- tically complete. Even though they have been originally introduced to solve the trajectory planning problem for holonomic robots, RRT-based algorithms have been ex- tended to tackle more complex problems, like for instance the optimal and constrained trajectory planning. Several solutions have been proposed in the last few years, includ- ing optimal and non-linear RRTs [2], Rapidly-exploring Random Graphs (RRG), and RRT* [5, 4]. Several ap- proaches have also been proposed to solve the optimal and constrained trajectory planning problem for arbitrary kino-dynamic systems, like for instance [3, 11, 12, 13]. A typical feature of all these randomized planning al- gorithms consists in sampling the nodes directly in the robot Configuration-Space, also known as Joint-Space. Unfortunately, Joint-Space randomized algorithms tend to output quite counterintuitive trajectories from the point of view of a human being. Given the fact that in construction sites humans and machines usually work side-by-side, and considering the results presented in [15], it is clear that the intuitiveness of the planned trajectories can have a strong impact on the comfort, the productivity and the level of safety perceived by human workers sharing their workspace with ACMs. In order to avoid these draw- backs, a possible solution is represented by addressing the motion planning problem directly at the Cartesian-Space level. We here propose “cart-RRT”, a Cartesian-Space randomized algorithm that allows the tree to be extended directly towards the goal, as it has been originally pro- posed in [6]. In cart-RRT edges are computed as linear segments in the Cartesian-Space and the corresponding joint positions profiles are determined using Inverse Kine- matics. Once the tree reaches the goal, a path shorten- ing procedure is performed to reduce the length of the