SIMPL is a simple and efficient multi-agent motion prediction baseline for autonomous driving. Unlike conventional agent-centric methods with high accuracy but repetitive computations and scene-centric methods with compromised accuracy and generalizability, SIMPL delivers real-time, accurate motion predictions for all relevant traffic participants. To improve accuracy and inference speed, SIMPL introduces a compact and efficient global feature fusion module that performs symmetric message passing, enabling the network to forecast future motion for all road users in a single feed-forward pass and mitigating accuracy loss caused by viewpoint shifting. Additionally, SIMPL uses Bernstein basis polynomials for continuous trajectory parameterization, allowing evaluations of states and their higher-order derivatives at any desired time point, which is valuable for downstream planning tasks. SIMPL achieves highly competitive performance on Argoverse 1 & 2 motion forecasting benchmarks compared to other state-of-the-art methods. Its lightweight design and low inference latency make it highly extensible and promising for real-world onboard deployment. The code is open-sourced at https://github.com/HKUST-Aerial-Robotics/SIMPL. The paper introduces SIMPL, a simple and efficient motion prediction baseline for autonomous driving. It addresses critical issues in multi-agent trajectory prediction for real-world onboard applications. SIMPL introduces an instance-centric scene representation followed by a symmetric fusion Transformer (SFT), enabling efficient trajectory forecasting for all agents in a single feed-forward pass while retaining accuracy and robustness. Compared to other recent works based on symmetric context fusion, the proposed SFT is simpler, more lightweight, and easier to implement, making it suitable for onboard deployment. SIMPL also introduces a novel parameterization method for predicted trajectories based on Bernstein basis polynomials (Bézier curves), ensuring smoothness and enabling effortless evaluation of exact states and their higher-order derivatives at any given time point. Empirical studies show that learning to forecast the control points of Bézier curves is more effective and numerically stable compared to estimating the coefficients of monomial basis polynomials. The proposed components are well integrated into a simple and efficient model. SIMPL is evaluated on two large-scale motion forecasting datasets and shows highly competitive performance compared to other state-of-the-art methods despite its streamlined design. SIMPL achieves efficient multi-agent trajectory prediction with fewer learnable parameters and lower inference latency without sacrificing quantitative performance, making it promising for real-world onboard deployment. The paper also highlights that SIMPL gains excellent extensibility as a strong baseline. The succinct architecture facilitates straightforward integration with recent advances in motion forecasting, offering opportunities for further enhancements in overall performance.SIMPL is a simple and efficient multi-agent motion prediction baseline for autonomous driving. Unlike conventional agent-centric methods with high accuracy but repetitive computations and scene-centric methods with compromised accuracy and generalizability, SIMPL delivers real-time, accurate motion predictions for all relevant traffic participants. To improve accuracy and inference speed, SIMPL introduces a compact and efficient global feature fusion module that performs symmetric message passing, enabling the network to forecast future motion for all road users in a single feed-forward pass and mitigating accuracy loss caused by viewpoint shifting. Additionally, SIMPL uses Bernstein basis polynomials for continuous trajectory parameterization, allowing evaluations of states and their higher-order derivatives at any desired time point, which is valuable for downstream planning tasks. SIMPL achieves highly competitive performance on Argoverse 1 & 2 motion forecasting benchmarks compared to other state-of-the-art methods. Its lightweight design and low inference latency make it highly extensible and promising for real-world onboard deployment. The code is open-sourced at https://github.com/HKUST-Aerial-Robotics/SIMPL. The paper introduces SIMPL, a simple and efficient motion prediction baseline for autonomous driving. It addresses critical issues in multi-agent trajectory prediction for real-world onboard applications. SIMPL introduces an instance-centric scene representation followed by a symmetric fusion Transformer (SFT), enabling efficient trajectory forecasting for all agents in a single feed-forward pass while retaining accuracy and robustness. Compared to other recent works based on symmetric context fusion, the proposed SFT is simpler, more lightweight, and easier to implement, making it suitable for onboard deployment. SIMPL also introduces a novel parameterization method for predicted trajectories based on Bernstein basis polynomials (Bézier curves), ensuring smoothness and enabling effortless evaluation of exact states and their higher-order derivatives at any given time point. Empirical studies show that learning to forecast the control points of Bézier curves is more effective and numerically stable compared to estimating the coefficients of monomial basis polynomials. The proposed components are well integrated into a simple and efficient model. SIMPL is evaluated on two large-scale motion forecasting datasets and shows highly competitive performance compared to other state-of-the-art methods despite its streamlined design. SIMPL achieves efficient multi-agent trajectory prediction with fewer learnable parameters and lower inference latency without sacrificing quantitative performance, making it promising for real-world onboard deployment. The paper also highlights that SIMPL gains excellent extensibility as a strong baseline. The succinct architecture facilitates straightforward integration with recent advances in motion forecasting, offering opportunities for further enhancements in overall performance.