Formal verification of complex robotic systems on resource-constrained platforms

Abstract

Software constitutes a major part of the development of robotic and autonomous systems and is critical to their successful deployment in our everyday life. Robotic software must thus run and perform as specified. Since most of these systems are used in a hard real-time context, the schedulability of their tasks is a crucial property. In this work, we propose to use formal methods to check whether the tasks of a robotic application are schedulable with respect to a given hardware platform. For this, we automatically translate functional components specified in GenoM into FIACRE, a formal language for timed systems. The generated models integrate realistic real-time schedulers based on the FCFS and the SJF cooperative policies. We use then the model-checker TINA to assert schedulability properties. We carry out experiments on a real robotic system, namely a quadcopter flight controller. We demonstrate that, on its actual hardware, schedulability properties can be formally expressed and verified. We give examples on how we can check other important behavioral and timed properties on the same synthesized models.

Recent Trends in Formal Validation and Verification of Autonomous Robots Software

“…Similarly, we can compute the maximum time it takes between a stop request sent to NAVIGATION and the writing of the zero speed on the robot HW controller by ROBOTDRIVER from the Cmd port of SAFETYPILOT . Given the proper model of scheduler also written in FIACRE, we can check than the number of core on the CPU is sufficient or not [Foughali et al, 2018a]. Overall, even if model checking techniques suffer from state space explosion, the results obtained here on fairly complex robotic experiments are still encouraging.…”

“…LTL and patterns for Fiacre, TCTL for UPPAAL) and how to interpret the results, still, it is a big step forward in providing V&V tools to roboticists. To formally validate the obtained model, the semantics of G en oM has been first specified in Timed Transition Systems and then transformed in Timed Automata with Urgency and Data (See [Foughali et al, 2018a] for more details and proofs). The point being that whoever specify and implement components in G en oM, gets all these equivalent formal models for free, and can run the various V&V tools associated to them.…”

Statistical Model Checking of Complex Robotic Systems

“…Comparison to our previous work: In our previous efforts to verify the quadcopter, model checking scaled only for the stationary flight, excluding the MANEUVER component [8,9]. This is the first work that verifies the navigation application, involving all the components, through sound and automatic bridging with UPPAAL-SMC.…”

On Reconciling Schedulability Analysis and Model Checking in Robotics

“…UPPAAL [7] and IM-ITATOR [6]). Also, we already have templates that translate robotic specifications to both Fiacre/TINA [12] and UPPAAL [15]. Exploring both TPN and UTA will help us conclude on which of these templates we need to extend with schedulers.…”

“…In previous work, we bridged the robotic framework G en oM3 (Sect. 3.1) with Fiacre/TINA [12,13] and UPPAAL [15] through templates. Now, we only extend the UPPAAL template (since the optimized method, Sect.…”