To improve the real-time performance and safety of intelligent bus lane-changing
and obstacle avoidance in complex road environments, this study proposes a
multi-objective optimization algorithm called LMCTS. L-MCTS integrates a
lane-changing benefit model, an LSTM network, and Monte Carlo Tree Search.
First, the NGSIM dataset was utilized to filter lane-changing intention points
and surrounding traffic flow information, and classification rules were
established to process lane-changing behaviors. Based on these decision
outcomes, a multi-objective trajectory planning method was designed, taking into
account factors such as comfort, safety, and smoothness. The proposed algorithm
was validated on the CARLA simulation platform and compared with traditional
MCTS and DP+QP algorithms. Results indicated that, in actual driving scenarios,
the safety evaluation of L-MCTS improved by 10.71% compared to MCTS and by
17.72% compared to DP+QP. Additionally, L-MCTS enhanced comfort by 4.94% over
MCTS and by 2.41% over DP+QP, significantly enhancing passenger comfort. The
average algorithm execution time was recorded at 6.21 ms, which represented a
14.12% improvement over MCTS, demonstrating excellent real-time performance.