Abstract
We address the problem of controlling a noisy differential drive mobile robot such that the probability of satisfying a specification given as a Bounded Linear Temporal Logic (BLTL) formula over a set of properties at the regions in the environment is maximized. We assume that the vehicle can determine its precise initial position in a known map of the environment. However, inspired by practical limitations, we assume that the vehicle is equipped with noisy actuators and, during its motion in the environment, it can only measure the angular velocity of its wheels using limited accuracy incremental encoders. Assuming the duration of the motion is finite, we map the measurements to a Markov Decision Process (MDP). We use recent results in Statistical Model Checking (SMC) to obtain an MDP control policy that maximizes the probability of satisfaction. We translate this policy to a vehicle feedback control strategy and show that the probability that the vehicle satisfies the specification in the environment is bounded from below by the probability of satisfying the specification on the MDP. We illustrate our method with simulations and experimental results.
1. Introduction
In classical robotics, motion planning often focuses on simple reachability: moving from point A to point B while avoiding obstacles. However, modern autonomous missions require more complex behaviors involving sequences, timing constraints, and conditional goals—for example, "pick up a load within 14 seconds and stay in the test area for at least 1 second." While Bounded Linear Temporal Logic (BLTL) provides a formal language for these missions, applying it to physical robots is challenging due to actuator noise, limited sensor resolution, and the computational complexity of mapping continuous robot dynamics to discrete logic states.
This paper presents a framework that bridges the gap between low-level noisy sensory inputs and high-level temporal logic specifications, providing a method to maximize the probability of mission success despite hardware imperfections.
2. Methodology
The proposed solution follows a multi-step approach to handle uncertainty and scalability:
2.1 MDP Modeling and Statistical Model Checking
The robot's motion and encoder readings are discretized into a Markov Decision Process (MDP). Each state represents a sequence of measured wheel rotations. Because the resulting state space is too large for exact solvers, the authors employ Statistical Model Checking (SMC). Specifically, a Control Policy Optimization algorithm samples paths to incrementally improve the control strategy, while Bayesian Interval Estimation (BIE) provides a confidence interval for the probability of satisfying the BLTL formula.
2.2 Conservative Uncertainty Bounds
To ensure the discrete model accurately reflects the continuous world, the robot's position is modeled as a "disc" of uncertainty that grows over time due to accumulated noise. A mission requirement is only considered "satisfied" in the model if the entire uncertainty disc meets the criteria. This conservative approach ensures that the success probability calculated on the MDP acts as a mathematical lower bound for the real-world success probability.
3. Experimental Setup
The framework was validated using a Dr. Robot x80Pro mobile robot. The authors first characterized the continuous probability density function for actuator noise by measuring actual wheel velocities against commanded inputs. During experimental trials, an OptiTrack motion capture system with eight cameras was used to reconstruct ground-truth trajectories for validation against the model's predictions.
4. Results and Discussion
The framework was tested in simulated environments with complex "pick-up, test, and drop-off" missions. In "Environment A," the MDP estimated a success probability of 0.664, while actual simulations yielded a significantly higher success rate of approximately 0.835. Similar results were observed in "Environment B," where the predicted 0.719 probability was outperformed by an actual 0.889 success rate.
These results confirm the "lower bound" theory: by being conservative in the uncertainty model, the robot performs better in practice than the model predicts. However, the gap between predicted and actual success suggests the model may be overly conservative, presenting an opportunity for future refinement.
5. Conclusion
This work demonstrates a scalable approach to controlling noisy robots under complex temporal constraints. By using sampling-based SMC instead of exhaustive state-space analysis, the problem becomes tractable for real-world applications. Future work will focus on reducing computation time—currently approximately 2.2 hours for policy generation—through parallel implementation and developing tighter uncertainty bounds for less conservative planning in dynamic or unknown environments.
Summary updated