Developing industrial robots which are safe, performant, robust and reliable over time is challenging, because their embedded distributed software system involves complex motions with force and torque control and anti-collision surveillance processes. Generating test trajectories which increase the chance to uncover potential failures or downtime is thus crucial to verify the reliability and performance of the robot before delivering it to its final users. Currently, these trajectories are manually created by test engineers, something that renders the process error-prone and time-consuming. In this paper, we present RobTest, a Constraint Programming approach for generating automatically maximal test trajectories for serial industrial robots. RobTest sequentially calls two constraint solvers: a solver over continuous domains to determine the reachability between configurations of the robot’s 3D-space, and a solver over finite domains to generate maximal-load test trajectories among a set of input points and obstacles of the 3D-space. RobTest is developed at ABB Robotics, a large robot manufacturing company, together with test engineers, who are preparing it for integration within the continuous testing process of the robots product-line. This paper reports on initial experimental results with three distinct solvers, namely Gecode, SICStus and Chuffed, where RobTest, has been shown to return near-optimal solutions for trajectories encounting for more than 80 input points and 60 obstacles in less than 5 min.