This paper addresses the challenges of motion planning for autonomous robots in dynamic environments, particularly the issue of local minima and the choice of sampling distribution in Model Predictive Path Integral (MPPI) control. The authors propose a novel approach called Biased-MPPI, which allows for arbitrary sampling distributions to enhance efficiency, robustness, and convergence. The key contributions include: 1. **Mathematical Derivation**: The paper provides a mathematical derivation of Biased-MPPI, showing how to incorporate biases in the sampling distribution to improve performance. 2. **Importance Sampling Scheme**: An efficient importance sampling scheme is presented, combining classical and learning-based ancillary controllers to take more informative samples. 3. **Control Fusion**: The method acts as a control fusion scheme, leveraging multiple underlying controllers to improve robustness to model uncertainties and local minima. 4. **Experimental Validation**: The effectiveness of Biased-MPPI is demonstrated through simulated and real-world experiments, including a rotary inverted pendulum and a multi-agent interaction-aware motion planning task. The results show that Biased-MPPI outperforms classical MPPI and other methods in terms of robustness, performance, and sample efficiency, while maintaining the ability to handle complex and dynamic environments.This paper addresses the challenges of motion planning for autonomous robots in dynamic environments, particularly the issue of local minima and the choice of sampling distribution in Model Predictive Path Integral (MPPI) control. The authors propose a novel approach called Biased-MPPI, which allows for arbitrary sampling distributions to enhance efficiency, robustness, and convergence. The key contributions include: 1. **Mathematical Derivation**: The paper provides a mathematical derivation of Biased-MPPI, showing how to incorporate biases in the sampling distribution to improve performance. 2. **Importance Sampling Scheme**: An efficient importance sampling scheme is presented, combining classical and learning-based ancillary controllers to take more informative samples. 3. **Control Fusion**: The method acts as a control fusion scheme, leveraging multiple underlying controllers to improve robustness to model uncertainties and local minima. 4. **Experimental Validation**: The effectiveness of Biased-MPPI is demonstrated through simulated and real-world experiments, including a rotary inverted pendulum and a multi-agent interaction-aware motion planning task. The results show that Biased-MPPI outperforms classical MPPI and other methods in terms of robustness, performance, and sample efficiency, while maintaining the ability to handle complex and dynamic environments.