Robust Trajectory Prediction for Mobile Robots via Minimum Error Entropy Criterion and Adaptive LSTM Networks.

Area of Science:

• Robotics
• Machine Learning
• Computer Vision

Background:

• Robot navigation relies on accurate trajectory prediction.
• Standard deep learning models often use Mean Squared Error (MSE), which is vulnerable to real-world noise like sensor glitches and occlusions.
• This fragility limits the reliability of robots in practical environments.

Purpose of the Study:

• To develop a robust trajectory prediction framework resilient to non-Gaussian impulsive noise.
• To improve the reliability of robot navigation systems in degraded sensing conditions.

Main Methods:

• Proposed MEE-LSTM, integrating Long Short-Term Memory networks with the Minimum Error Entropy (MEE) criterion.
• Utilized Renyi's quadratic entropy minimization for inherent gradient clipping against outliers.
• Introduced Silverman-based Adaptive Annealing (SAA) to manage kernel bandwidth for stable information theoretic learning.

Main Results:

• MEE-LSTM demonstrated competitive accuracy on clean datasets and superior resilience in noisy environments.
• Under 20% impulsive noise, MEE-LSTM achieved an Average Displacement Error (ADE) of ≈0.51 m, while MSE baselines degraded significantly (ADE > 2.1 m).
• This represents a 75.7% improvement in robustness compared to MSE-based methods.

Conclusions:

• MEE-LSTM offers a statistically grounded approach for reliable trajectory prediction in challenging robotic perception scenarios.
• The proposed method significantly enhances robot navigation safety and robustness against sensor noise.
• This work paves the way for more dependable autonomous systems in unpredictable environments.