In advancing pedestrian safety in autonomous vehicle navigation, this contemporary research integrates Long Short-Term Memory (LSTM) networks that are sophisticated variants of Recurrent Neural Networks (RNN) with Model Predictive Control (MPC) and Signal Temporal Logic (STL). It focuses on LSTM's capability for predicting pedestrian trajectories, leveraging its advanced data processing ability to capture complex temporal dynamics and patterns in pedestrian movement. A novel aspect of this work is the incorporation of conformal prediction, improving the reliability of LSTM-based pedestrian trajectory forecasts. This predictive framework is integrated into an MPC system, which is further refined by STL to rigorously define safety constraints. Empirical simulations demonstrate the autonomous system's ability to adapt to real-time pedestrian movements, maintaining adherence to stringent safety specifications. Through the use of intricate techniques such as STL and conformal prediction, the research offers a robust and dynamic solution for pedestrian safety in urban environments. (Please note the pseudocode is available in the report)
