control-based motion generation

This work addresses a key limitation of existing training-free controllable motion generation methods, which are restricted to differentiable objective-type constraints and struggle with non-differentiable, sparse, or black-box criterion-type constraints. To overcome this, the authors propose the Motion-Inference-as-Control (MIaC) framework, which formulates diffusion-based motion generation as a stochastic control problem. MIaC is the first approach to unify both objective-type and criterion-type heterogeneous constraints without requiring retraining. It achieves dynamic conflict resolution among multiple constraints through a stepwise control policy, gradient-free constraint guidance, and an adaptive coordination mechanism. Experiments demonstrate that the method significantly improves motion quality and constraint satisfaction across diverse and complex constraint scenarios.

This study addresses the computational bottleneck associated with dynamically constrained sampling of state spaces in feedback control and planning. To overcome this challenge, the work reformulates control as a dynamically constrained sampling problem, establishing a mapping framework that bridges controllability, optimal control theory, and generative modeling. Specifically, it integrates flow matching, normalizing flows, and denoising diffusion techniques to guide system evolution toward target states or distributions. The proposed approach enables efficient reachable set sampling and precise trajectory planning while unifying control-theoretic and generative-modeling paradigms. Furthermore, the authors provide an accessible open-source tutorial to facilitate practical adoption by the research community.

Sample-based motion planners (SBMPs) suffer from low efficiency and slow convergence in complex environments due to uniform sampling, which wastes computational resources on low-value regions. Method: This paper proposes a theoretically grounded non-uniform sampling strategy that integrates conformal prediction—introduced to motion planning for the first time—to provide distribution-free, confidence-level–controlled guarantees that the optimal solution lies within certified high-probability sampling regions. Leveraging heuristic path predictors (e.g., A* or vision-language models), the method generates an initial trajectory and quantifies its epistemic uncertainty to identify high-value sampling zones. Contribution/Results: The approach significantly accelerates feasible path discovery while ensuring theoretical validity of sampling coverage. Extensive experiments demonstrate superior generalization and robustness over state-of-the-art baselines, particularly in previously unseen environments, validating both computational efficiency and reliability under uncertainty.

This work addresses the limited exploration capability and difficulty in converging to global optima inherent in sampling-based controllers for path planning. To overcome these challenges, the authors propose integrating motion primitives into the Model Predictive Path Integral (MPPI) framework. By fusing motion primitive-guided structured sampling with perturbed control sequences within the real-time optimization loop, the method substantially enhances exploration efficiency and global optimality in the control space while preserving MPPI’s intrinsic fast response characteristics. Evaluations on quadrotor obstacle navigation tasks demonstrate that the proposed algorithm significantly improves both exploratory behavior and real-time performance, thereby validating its superiority over conventional approaches.

To address control chattering and divergence of sampling-based Model Predictive Path Integral (MPPI) control in nonlinear systems—particularly under dynamic conditions due to stochastic sampling—this paper proposes an intrinsically smooth MPPI framework. Methodologically, it introduces: (1) an input-lifting strategy that embeds control inputs into a high-dimensional differentiable manifold to enhance action-space continuity; and (2) an information-theoretic action cost function that implicitly enforces temporal smoothness of the control sequence without external filtering or post-processing. The formulation rigorously preserves MPPI’s information-theoretic foundation and compatibility with non-affine dynamics. Evaluated on swing-up of an inverted pendulum and neural-network-modeled autonomous driving, the method significantly suppresses chattering and improves closed-loop stability and convergence robustness compared to baseline MPPI augmented with moving-average smoothing.

本文针对高维不稳定系统的控制问题，提出了一种结合反馈策略的混合采样方法（FS-MPC），提高了样本效率和控制性能。

本文提出了一种在线可达性感知的采样基运动规划方法，通过快速区间管道计算可达集近似，无需预计算步骤，减少了99%以上的安全违规。

本文针对基于采样的运动规划算法中δ-相似轨迹假设不成立导致的问题，提出了一种在考虑‘拥挤排除’情况下仍能实现渐近近似最优性的方法。

Sampling-based trajectory optimization is highly sensitive to initial solutions and exhibits limited exploration capability in complex environments, often converging to suboptimal local minima. To address this, this work proposes the UGE-TO algorithm, which, for the first time, explicitly models trajectory uncertainty in configuration space as a probability distribution induced by uncertainty ellipsoids. By incorporating the effects of system dynamics and action selection, the method enforces distributional separation via the Hellinger distance to generate diverse trajectory samples. Integrated into a model predictive control framework as UGE-MPC, the approach achieves a 72.1% faster convergence rate in obstacle-free environments and, in cluttered settings, converges 66% faster with a 6.7% higher success rate under the same sampling budget. Both simulation and real-world experiments validate its superior performance.

This study addresses the limitations of Model Predictive Control (MPC), which relies on manually designed cost functions, and existing empowerment-based control methods that require computationally expensive second-order derivatives while decoupling exploration from execution. To overcome these challenges, this work proposes a unified framework that integrates empowerment objectives and task behaviors within a single policy. By combining information-theoretic empowerment estimation with sampling-based MPC, the approach requires only first-order dynamics derivatives and can be optimized via standard MPC solvers, intrinsically unifying exploration and execution. This method eliminates virtual probe interference, significantly reduces computational complexity, and remains compatible with mainstream MPC paradigms. Experimental results demonstrate that the proposed approach effectively balances empowered exploration and task execution in classical control tasks, achieving success rates comparable to or exceeding those of single-objective optimization, thereby bridging standard MPC and intrinsic motivation-based control.