Wanxin Jin

Yixuan Wang, Simon Sinong Zhan, Ruochen Jiao et al.

It is quite challenging to ensure the safety of reinforcement learning (RL) agents in an unknown and stochastic environment under hard constraints that require the system state not to reach certain specified unsafe regions. Many popular safe RL methods such as those based on the Constrained Markov Decision Process (CMDP) paradigm formulate safety violations in a cost function and try to constrain the expectation of cumulative cost under a threshold. However, it is often difficult to effectively capture and enforce hard reachability-based safety constraints indirectly with such constraints on safety violation costs. In this work, we leverage the notion of barrier function to explicitly encode the hard safety constraints, and given that the environment is unknown, relax them to our design of \emph{generative-model-based soft barrier functions}. Based on such soft barriers, we propose a safe RL approach that can jointly learn the environment and optimize the control policy, while effectively avoiding unsafe regions with safety probability optimization. Experiments on a set of examples demonstrate that our approach can effectively enforce hard safety constraints and significantly outperform CMDP-based baseline methods in system safe rate measured via simulations.

Wen Yang, Zhixian Xie, Yiting Wang et al.

Real-time tracking of previously unseen, highly dynamic objects in contact-rich scenes, such as during dexterous in-hand manipulation, remains a major challenge. Pure vision-based approaches often fail under heavy occlusions due to frequent contact interactions and motion blur caused by abrupt impacts. We propose Twintrack, a physics-aware perception system that enables robust, real-time 6-DoF pose tracking of unknown dynamic objects in contact-rich scenes by leveraging contact physics cues. At its core, Twintrack integrates Real2Sim and Sim2Real. Real2Sim combines vision and contact physics to jointly estimate object geometry and physical properties: an initial reconstruction is obtained from vision, then refined by learning a geometry residual and simultaneously estimating physical parameters (e.g., mass, inertia, and friction) based on contact dynamics consistency. Sim2Real achieves robust pose estimation by adaptively fusing a visual tracker with predictions from the updated contact dynamics. Twintrack is implemented on a GPU-accelerated, customized MJX engine to guarantee real-time performance. We evaluate our method on two contact-rich scenarios: object falling with environmental contacts and multi-fingered in-hand manipulation. Results show that, compared to baselines, Twintrack delivers significantly more robust, accurate, and real-time tracking in these challenging settings, with tracking speeds above 20 Hz. Project page: https://irislab.tech/TwinTrack-webpage/

Yizhi Zhou, Wanxin Jin, Xuan Wang

This paper develops a Distributed Differentiable Dynamic Game (D3G) framework, which can efficiently solve the forward and inverse problems in multi-robot coordination. We formulate multi-robot coordination as a dynamic game, where the behavior of a robot is dictated by its own dynamics and objective that also depends on others' behavior. In the forward problem, D3G enables all robots collaboratively to seek the Nash equilibrium of the game in a distributed manner, by developing a distributed shooting-based Nash solver. In the inverse problem, where each robot aims to find (learn) its objective (and dynamics) parameters to mimic given coordination demonstrations, D3G proposes a differentiation solver based on Differential Pontryagin's Maximum Principle, which allows each robot to update its parameters in a distributed and coordinated manner. We test the D3G in simulation with two types of robots given different task configurations. The results demonstrate the effectiveness of D3G for solving both forward and inverse problems in comparison with existing methods.

Ni Dang, Tao Shi, Zengjie Zhang et al.

The driving style of an Autonomous Vehicle (AV) refers to how it behaves and interacts with other AVs. In a multi-vehicle autonomous driving system, an AV capable of identifying the driving styles of its nearby AVs can reliably evaluate the risk of collisions and make more reasonable driving decisions. However, there has not been a consistent definition of driving styles for an AV in the literature, although it is considered that the driving style is encoded in the AV's trajectories and can be identified using Maximum Entropy Inverse Reinforcement Learning (ME-IRL) methods as a cost function. Nevertheless, an important indicator of the driving style, i.e., how an AV reacts to its nearby AVs, is not fully incorporated in the feature design of previous ME-IRL methods. In this paper, we describe the driving style as a cost function of a series of weighted features. We design additional novel features to capture the AV's reaction-aware characteristics. Then, we identify the driving styles from the demonstration trajectories generated by the Stochastic Model Predictive Control (SMPC) using a modified ME-IRL method with our newly proposed features. The proposed method is validated using MATLAB simulation and an off-the-shelf experiment.

Shenao Zhang, Sirui Zheng, Shuqi Ke et al.

Reinforcement learning (RL) has become the de facto standard practice for sequential decision-making problems by improving future acting policies with feedback. However, RL algorithms may require extensive trial-and-error interactions to collect useful feedback for improvement. On the other hand, recent developments in large language models (LLMs) have showcased impressive capabilities in language understanding and generation, yet they fall short in exploration and self-improvement capabilities for planning tasks, lacking the ability to autonomously refine their responses based on feedback. Therefore, in this paper, we study how the policy prior provided by the LLM can enhance the sample efficiency of RL algorithms. Specifically, we develop an algorithm named LINVIT that incorporates LLM guidance as a regularization factor in value-based RL, leading to significant reductions in the amount of data needed for learning, particularly when the difference between the ideal policy and the LLM-informed policy is small, which suggests that the initial policy is close to optimal, reducing the need for further exploration. Additionally, we present a practical algorithm SLINVIT that simplifies the construction of the value function and employs subgoals to reduce the search complexity. Our experiments across three interactive environments ALFWorld, InterCode, and BlocksWorld demonstrate that our method achieves state-of-the-art success rates and also surpasses previous RL and LLM approaches in terms of sample efficiency. Our code is available at https://github.com/agentification/Language-Integrated-VI.

Meizhong Wang, Wanxin Jin, Kun Cao et al.

Developing world models that understand complex physical interactions is essential for advancing robotic planning and simulation.However, existing methods often struggle to accurately model the environment under conditions of data scarcity and complex contact-rich dynamic motion.To address these challenges, we propose ContactGaussian-WM, a differentiable physics-grounded rigid-body world model capable of learning intricate physical laws directly from sparse and contact-rich video sequences.Our framework consists of two core components: (1) a unified Gaussian representation for both visual appearance and collision geometry, and (2) an end-to-end differentiable learning framework that differentiates through a closed-form physics engine to infer physical properties from sparse visual observations.Extensive simulations and real-world evaluations demonstrate that ContactGaussian-WM outperforms state-of-the-art methods in learning complex scenarios, exhibiting robust generalization capabilities.Furthermore, we showcase the practical utility of our framework in downstream applications, including data synthesis and real-time MPC.

Chetan Borse, Zhixian Xie, Wei-Cheng Huang et al.

Physics simulation for contact-rich robotics is often bottlenecked by contact resolution: mainstream engines enforce non-penetration and Coulomb friction via complementarity constraints or constrained optimization, requiring per-step iterative solves whose cost grows superlinearly with contact density. We present ComFree-Sim, a GPU-parallelized analytical contact physics engine built on complementarity-free contact modeling. ComFree-Sim computes contact impulses in closed form via an impedance-style prediction--correction update in the dual cone of Coulomb friction. Contact computation decouples across contact pairs and becomes separable across cone facets, mapping naturally to GPU kernels and yielding near-linear runtime scaling with the number of contacts. We further extend the formulation to a unified 6D contact model capturing tangential, torsional, and rolling friction, and introduce a practical dual-cone impedance heuristic. ComFree-Sim is implemented in Warp and exposed through a MuJoCo-compatible interface as a drop-in backend alternative to MuJoCo Warp (MJWarp). Experiments benchmark penetration, friction behaviors, stability, and simulation runtime scaling against MJWarp, demonstrating near-linear scaling and 2--3 times higher throughput in dense contact scenes with comparable physical fidelity. We deploy ComFree-Sim in real-time MPC for in-hand dexterous manipulation on a real-world multi-fingered LEAP hand and in dynamics-aware motion retargeting, demonstrating that low-latency simulation yields higher closed-loop success rates and enables practical high-frequency control in contact-rich tasks.

Harsh Mahesheka, Zhixian Xie, Zhaoran Wang et al.

Learning from Demonstrations, particularly from biological experts like humans and animals, often encounters significant data acquisition challenges. While recent approaches leverage internet videos for learning, they require complex, task-specific pipelines to extract and retarget motion data for the agent. In this work, we introduce a language-model-assisted bi-level programming framework that enables a reinforcement learning agent to directly learn its reward from internet videos, bypassing dedicated data preparation. The framework includes two levels: an upper level where a vision-language model (VLM) provides feedback by comparing the learner's behavior with expert videos, and a lower level where a large language model (LLM) translates this feedback into reward updates. The VLM and LLM collaborate within this bi-level framework, using a "chain rule" approach to derive a valid search direction for reward learning. We validate the method for reward learning from YouTube videos, and the results have shown that the proposed method enables efficient reward design from expert videos of biological agents for complex behavior synthesis.

Olivia Dry, Timothy L. Molloy, Wanxin Jin et al.

We propose Zeroth-Order Random Matrix Search for Learning from Demonstrations (ZORMS-LfD). ZORMS-LfD enables the costs, constraints, and dynamics of constrained optimal control problems, in both continuous and discrete time, to be learned from expert demonstrations without requiring smoothness of the learning-loss landscape. In contrast, existing state-of-the-art first-order methods require the existence and computation of gradients of the costs, constraints, dynamics, and learning loss with respect to states, controls and/or parameters. Most existing methods are also tailored to discrete time, with constrained problems in continuous time receiving only cursory attention. We demonstrate that ZORMS-LfD matches or surpasses the performance of state-of-the-art methods in terms of both learning loss and compute time across a variety of benchmark problems. On unconstrained continuous-time benchmark problems, ZORMS-LfD achieves similar loss performance to state-of-the-art first-order methods with an over $80$\% reduction in compute time. On constrained continuous-time benchmark problems where there is no specialized state-of-the-art method, ZORMS-LfD is shown to outperform the commonly used gradient-free Nelder-Mead optimization method.

Zhixian Xie, Haode Zhang, Yizhe Feng et al.

Reward design in reinforcement learning and optimal control is challenging. Preference-based alignment addresses this by enabling agents to learn rewards from ranked trajectory pairs provided by humans. However, existing methods often struggle from poor robustness to unknown false human preferences. In this work, we propose a robust and efficient reward alignment method based on a novel and geometrically interpretable perspective: hypothesis space batched cutting. Our method iteratively refines the reward hypothesis space through "cuts" based on batches of human preferences. Within each batch, human preferences, queried based on disagreement, are grouped using a voting function to determine the appropriate cut, ensuring a bounded human query complexity. To handle unknown erroneous preferences, we introduce a conservative cutting method within each batch, preventing erroneous human preferences from making overly aggressive cuts to the hypothesis space. This guarantees provable robustness against false preferences, while eliminating the need to explicitly identify them. We evaluate our method in a model predictive control setting across diverse tasks. The results demonstrate that our framework achieves comparable or superior performance to state-of-the-art methods in error-free settings while significantly outperforming existing methods when handling a high percentage of erroneous human preferences.

Wanxin Jin, Alp Aydinoglu, Mathew Halm et al.

This paper investigates the learning, or system identification, of a class of piecewise-affine dynamical systems known as linear complementarity systems (LCSs). We propose a violation-based loss which enables efficient learning of the LCS parameterization, without prior knowledge of the hybrid mode boundaries, using gradient-based methods. The proposed violation-based loss incorporates both dynamics prediction loss and a novel complementarity - violation loss. We show several properties attained by this loss formulation, including its differentiability, the efficient computation of first- and second-order derivatives, and its relationship to the traditional prediction loss, which strictly enforces complementarity. We apply this violation-based loss formulation to learn LCSs with tens of thousands of (potentially stiff) hybrid modes. The results demonstrate a state-of-the-art ability to identify piecewise-affine dynamics, outperforming methods which must differentiate through non-smooth linear complementarity problems.

Yunxi Tang, Xiangyu Chu, Wanxin Jin et al.

Trajectory optimization has been used extensively in robotic systems. In particular, iterative Linear Quadratic Regulator (iLQR) has performed well as an off-line planner and online nonlinear model predictive control solver, with a lower computational cost. However, standard iLQR cannot handle any constraints or perform reasonable initialization of a state trajectory. In this paper, we propose a hybrid constrained iLQR variant with a multiple-shooting framework to incorporate general inequality constraints and infeasible states initialization. The main technical contributions are twofold: 1) In addition to inheriting the simplicity of the initialization in multiple-shooting settings, a two-stage framework is developed to deal with state and/or control constraints robustly without loss of the linear feedback term of iLQR. Such a hybrid strategy offers fast convergence of constraint satisfaction. 2) An improved globalization strategy is proposed to exploit the coupled effects between line-searching and regularization, which is able to enhance the numerical robustness of the constrained iLQR approaches. Our approach is tested on various constrained trajectory optimization problems and outperforms the commonly-used collocation and shooting methods.

Wanxin Jin, Shaoshuai Mou, George J. Pappas

We propose a Safe Pontryagin Differentiable Programming (Safe PDP) methodology, which establishes a theoretical and algorithmic framework to solve a broad class of safety-critical learning and control tasks -- problems that require the guarantee of safety constraint satisfaction at any stage of the learning and control progress. In the spirit of interior-point methods, Safe PDP handles different types of system constraints on states and inputs by incorporating them into the cost or loss through barrier functions. We prove three fundamentals of the proposed Safe PDP: first, both the solution and its gradient in the backward pass can be approximated by solving their more efficient unconstrained counterparts; second, the approximation for both the solution and its gradient can be controlled for arbitrary accuracy by a barrier parameter; and third, importantly, all intermediate results throughout the approximation and optimization strictly respect the constraints, thus guaranteeing safety throughout the entire learning and control process. We demonstrate the capabilities of Safe PDP in solving various safety-critical tasks, including safe policy optimization, safe motion planning, and learning MPCs from demonstrations, on different challenging systems such as 6-DoF maneuvering quadrotor and 6-DoF rocket powered landing.

Wanxin Jin, Todd D. Murphey, Zehui Lu et al.

This paper proposes a novel approach that enables a robot to learn an objective function incrementally from human directional corrections. Existing methods learn from human magnitude corrections; since a human needs to carefully choose the magnitude of each correction, those methods can easily lead to over-corrections and learning inefficiency. The proposed method only requires human directional corrections -- corrections that only indicate the direction of an input change without indicating its magnitude. We only assume that each correction, regardless of its magnitude, points in a direction that improves the robot's current motion relative to an unknown objective function. The allowable corrections satisfying this assumption account for half of the input space, as opposed to the magnitude corrections which have to lie in a shrinking level set. For each directional correction, the proposed method updates the estimate of the objective function based on a cutting plane method, which has a geometric interpretation. We have established theoretical results to show the convergence of the learning process. The proposed method has been tested in numerical examples, a user study on two human-robot games, and a real-world quadrotor experiment. The results confirm the convergence of the proposed method and further show that the method is significantly more effective (higher success rate), efficient/effortless (less human corrections needed), and potentially more accessible (fewer early wasted trials) than the state-of-the-art robot learning frameworks.

Zihao Liang, Wanxin Jin, Shaoshuai Mou

This paper proposes an inverse optimal control method which enables a robot to incrementally learn a control objective function from a collection of trajectory segments. By saying incrementally, it means that the collection of trajectory segments is enlarged because additional segments are provided as time evolves. The unknown objective function is parameterized as a weighted sum of features with unknown weights. Each trajectory segment is a small snippet of optimal trajectory. The proposed method shows that each trajectory segment, if informative, can pose a linear constraint to the unknown weights, thus, the objective function can be learned by incrementally incorporating all informative segments. Effectiveness of the method is shown on a simulated 2-link robot arm and a 6-DoF maneuvering quadrotor system, in each of which only small demonstration segments are available.

Wanxin Jin, Todd D. Murphey, Dana Kulić et al.

This paper develops the method of Continuous Pontryagin Differentiable Programming (Continuous PDP), which enables a robot to learn an objective function from a few sparsely demonstrated keyframes. The keyframes, labeled with some time stamps, are the desired task-space outputs, which a robot is expected to follow sequentially. The time stamps of the keyframes can be different from the time of the robot's actual execution. The method jointly finds an objective function and a time-warping function such that the robot's resulting trajectory sequentially follows the keyframes with minimal discrepancy loss. The Continuous PDP minimizes the discrepancy loss using projected gradient descent, by efficiently solving the gradient of the robot trajectory with respect to the unknown parameters. The method is first evaluated on a simulated robot arm and then applied to a 6-DoF quadrotor to learn an objective function for motion planning in unmodeled environments. The results show the efficiency of the method, its ability to handle time misalignment between keyframes and robot execution, and the generalization of objective learning into unseen motion conditions.

Wanxin Jin, Zhaoran Wang, Zhuoran Yang et al.

This paper develops an approach to learn a policy of a dynamical system that is guaranteed to be both provably safe and goal-reaching. Here, the safety means that a policy must not drive the state of the system to any unsafe region, while the goal-reaching requires the trajectory of the controlled system asymptotically converges to a goal region (a generalization of stability). We obtain the safe and goal-reaching policy by jointly learning two additional certificate functions: a barrier function that guarantees the safety and a developed Lyapunov-like function to fulfill the goal-reaching requirement, both of which are represented by neural networks. We show the effectiveness of the method to learn both safe and goal-reaching policies on various systems, including pendulums, cart-poles, and UAVs.

Wanxin Jin, Zhaoran Wang, Zhuoran Yang et al.

This paper develops a Pontryagin Differentiable Programming (PDP) methodology, which establishes a unified framework to solve a broad class of learning and control tasks. The PDP distinguishes from existing methods by two novel techniques: first, we differentiate through Pontryagin's Maximum Principle, and this allows to obtain the analytical derivative of a trajectory with respect to tunable parameters within an optimal control system, enabling end-to-end learning of dynamics, policies, or/and control objective functions; and second, we propose an auxiliary control system in the backward pass of the PDP framework, and the output of this auxiliary control system is the analytical derivative of the original system's trajectory with respect to the parameters, which can be iteratively solved using standard control tools. We investigate three learning modes of the PDP: inverse reinforcement learning, system identification, and control/planning. We demonstrate the capability of the PDP in each learning mode on different high-dimensional systems, including multi-link robot arm, 6-DoF maneuvering quadrotor, and 6-DoF rocket powered landing.

Wanxin Jin, Dana Kulić, Shaoshuai Mou et al.

This article develops a methodology that enables learning an objective function of an optimal control system from incomplete trajectory observations. The objective function is assumed to be a weighted sum of features (or basis functions) with unknown weights, and the observed data is a segment of a trajectory of system states and inputs. The proposed technique introduces the concept of the recovery matrix to establish the relationship between any available segment of the trajectory and the weights of given candidate features. The rank of the recovery matrix indicates whether a subset of relevant features can be found among the candidate features and the corresponding weights can be learned from the segment data. The recovery matrix can be obtained iteratively and its rank non-decreasing property shows that additional observations may contribute to the objective learning. Based on the recovery matrix, a method for using incomplete trajectory observations to learn the weights of selected features is established, and an incremental inverse optimal control algorithm is developed by automatically finding the minimal required observation. The effectiveness of the proposed method is demonstrated on a linear quadratic regulator system and a simulated robot manipulator.