Junhong Xu

Junhong Xu, Durgakant Pushp, Kai Yin et al.

When robots share the same workspace with other intelligent agents (e.g., other robots or humans), they must be able to reason about the behaviors of their neighboring agents while accomplishing the designated tasks. In practice, frequently, agents do not exhibit absolutely rational behavior due to their limited computational resources. Thus, predicting the optimal agent behaviors is undesirable (because it demands prohibitive computational resources) and undesirable (because the prediction may be wrong). Motivated by this observation, we remove the assumption of perfectly rational agents and propose incorporating the concept of bounded rationality from an information-theoretic view into the game-theoretic framework. This allows the robots to reason other agents' sub-optimal behaviors and act accordingly under their computational constraints. Specifically, bounded rationality directly models the agent's information processing ability, which is represented as the KL-divergence between nominal and optimized stochastic policies, and the solution to the bounded-optimal policy can be obtained by an efficient importance sampling approach. Using both simulated and real-world experiments in multi-robot navigation tasks, we demonstrate that the resulting framework allows the robots to reason about different levels of rational behaviors of other agents and compute a reasonable strategy under its computational constraint.

Youwei Yu, Junhong Xu, Lantao Liu

Model-free reinforcement learning has emerged as a powerful method for developing robust robot control policies capable of navigating through complex and unstructured terrains. The effectiveness of these methods hinges on two essential elements: (1) the use of massively parallel physics simulations to expedite policy training, and (2) an environment generator tasked with crafting sufficiently challenging yet attainable terrains to facilitate continuous policy improvement. Existing methods of environment generation often rely on heuristics constrained by a set of parameters, limiting the diversity and realism. In this work, we introduce the Adaptive Diffusion Terrain Generator (ADTG), a novel method that leverages Denoising Diffusion Probabilistic Models to dynamically expand existing training environments by adding more diverse and complex terrains adaptive to the current policy. ADTG guides the diffusion model's generation process through initial noise optimization, blending noise-corrupted terrains from existing training environments weighted by the policy's performance in each corresponding environment. By manipulating the noise corruption level, ADTG seamlessly transitions between generating similar terrains for policy fine-tuning and novel ones to expand training diversity. Our experiments show that the policy trained by ADTG outperforms both procedural generated and natural environments, along with popular navigation methods.

Alejandro Murillo-Gonzalez, Junhong Xu, Lantao Liu

Structural causal models describe how the components of a robotic system interact. They provide both structural and functional information about the relationships that are present in the system. The structural information outlines the variables among which there is interaction. The functional information describes how such interactions work, via equations or learned models. In this paper we find that learning the functional relationships while accounting for the uncertainty about the structural information leads to more robust dynamics models which improves downstream planning, while using significantly lower computational resources. This in contrast with common model-learning methods that ignore the causal structure and fail to leverage the sparsity of interactions in robotic systems. We achieve this by estimating a causal structure distribution that is used to sample causal graphs that inform the latent-space representations in an encoder-multidecoder probabilistic model. We show that our model can be used to learn the dynamics of a robot, which together with a sampling-based planner can be used to perform new tasks in novel environments, provided an objective function for the new requirement is available. We validate our method using manipulators and mobile robots in both simulation and the real-world. Additionally, we validate the learned dynamics' adaptability and increased robustness to corrupted inputs and changes in the environment, which is highly desirable in challenging real-world robotics scenarios. Video: https://youtu.be/X6k5t7OOnNc.

Jonathan Booher, Khashayar Rohanimanesh, Junhong Xu et al.

Modern approaches to autonomous driving rely heavily on learned components trained with large amounts of human driving data via imitation learning. However, these methods require large amounts of expensive data collection and even then face challenges with safely handling long-tail scenarios and compounding errors over time. At the same time, pure Reinforcement Learning (RL) methods can fail to learn performant policies in sparse, constrained, and challenging-to-define reward settings such as autonomous driving. Both of these challenges make deploying purely cloned or pure RL policies in safety critical applications such as autonomous vehicles challenging. In this paper we propose Combining IMitation and Reinforcement Learning (CIMRL) approach - a safe reinforcement learning framework that enables training driving policies in simulation through leveraging imitative motion priors and safety constraints. CIMRL does not require extensive reward specification and improves on the closed loop behavior of pure cloning methods. By combining RL and imitation, we demonstrate that our method achieves state-of-the-art results in closed loop simulation and real world driving benchmarks.

Junhong Xu, Kai Yin, Zheng Chen et al.

We propose a diffusion approximation method to the continuous-state Markov Decision Processes (MDPs) that can be utilized to address autonomous navigation and control in unstructured off-road environments. In contrast to most decision-theoretic planning frameworks that assume fully known state transition models, we design a method that eliminates such a strong assumption that is often extremely difficult to engineer in reality. We first take the second-order Taylor expansion of the value function. The Bellman optimality equation is then approximated by a partial differential equation, which only relies on the first and second moments of the transition model. By combining the kernel representation of the value function, we design an efficient policy iteration algorithm whose policy evaluation step can be represented as a linear system of equations characterized by a finite set of supporting states. We first validate the proposed method through extensive simulations in 2D obstacle avoidance and 2.5D terrain navigation problems. The results show that the proposed approach leads to a much superior performance over several baselines. We then develop a system that integrates our decision-making framework with onboard perception and conduct real-world experiments in both cluttered indoor and unstructured outdoor environments. The results from the physical systems further demonstrate the applicability of our method in challenging real-world environments.

Junhong Xu, Kai Yin, Lantao Liu

We investigate the autonomous navigation of a mobile robot in the presence of other moving vehicles under time-varying uncertain environmental disturbances. We first predict the future state distributions of other vehicles to account for their uncertain behaviors affected by the time-varying disturbances. We then construct a dynamic-obstacle-aware reachable space that contains states with high probabilities to be reached by the robot, within which the optimal policy is searched. Since, in general, the dynamics of both the vehicle and the environmental disturbances are nonlinear, we utilize a nonlinear Gaussian filter -- the unscented transform -- to approximate the future state distributions. Finally, the forward reachable space computation and backward policy search are iterated until convergence. Extensive simulation evaluations have revealed significant advantages of this proposed method in terms of computation time, decision accuracy, and planning reliability.

Junhong Xu, Kai Yin, Lantao Liu

We propose a principled kernel-based policy iteration algorithm to solve the continuous-state Markov Decision Processes (MDPs). In contrast to most decision-theoretic planning frameworks, which assume fully known state transition models, we design a method that eliminates such a strong assumption, which is oftentimes extremely difficult to engineer in reality. To achieve this, we first apply the second-order Taylor expansion of the value function. The Bellman optimality equation is then approximated by a partial differential equation, which only relies on the first and second moments of the transition model. By combining the kernel representation of value function, we then design an efficient policy iteration algorithm whose policy evaluation step can be represented as a linear system of equations characterized by a finite set of supporting states. We have validated the proposed method through extensive simulations in both simplified and realistic planning scenarios, and the experiments show that our proposed approach leads to a much superior performance over several baseline methods.

Junhong Xu, Kai Yin, Lantao Liu

We propose a solution to a time-varying variant of Markov Decision Processes which can be used to address decision-theoretic planning problems for autonomous systems operating in unstructured outdoor environments. We explore the time variability property of the planning stochasticity and investigate the state reachability, based on which we then develop an efficient iterative method that offers a good trade-off between solution optimality and time complexity. The reachability space is constructed by analyzing the means and variances of states' reaching time in the future. We validate our algorithm through extensive simulations using ocean data, and the results show that our method achieves a great performance in terms of both solution quality and computing time.

Junhong Xu, Kai Yin, Lantao Liu

Motion planning under uncertainty for an autonomous system can be formulated as a Markov Decision Process with a continuous state space. In this paper, we propose a novel solution to this decision-theoretic planning problem that directly obtains the continuous value function with only the first and second moments of the transition probabilities, alleviating the requirement for an explicit transition model in the literature. We achieve this by expressing the value function as a linear combination of basis functions and approximating the Bellman equation by a partial differential equation, where the value function can be naturally constructed using a finite element method. We have validated our approach via extensive simulations, and the evaluations reveal that to baseline methods, our solution leads to in terms of path smoothness, travel distance, and time costs.

Junhong Xu, Qiwei Liu, Hanqing Guo et al.

Deep imitation learning enables robots to learn from expert demonstrations to perform tasks such as lane following or obstacle avoidance. However, in the traditional imitation learning framework, one model only learns one task, and thus it lacks of the capability to support a robot to perform various different navigation tasks with one model in indoor environments. This paper proposes a new framework, Shared Multi-headed Imitation Learning(SMIL), that allows a robot to perform multiple tasks with one model without switching among different models. We model each task as a sub-policy and design a multi-headed policy to learn the shared information among related tasks by summing up activations from all sub-policies. Compared to single or non-shared multi-headed policies, this framework is able to leverage correlated information among tasks to increase performance.We have implemented this framework using a robot based on NVIDIA TX2 and performed extensive experiments in indoor environments with different baseline solutions. The results demonstrate that SMIL has doubled the performance over nonshared multi-headed policy.

Junhong Xu, Shangyue Zhu, Hanqing Guo et al.

Imitation learning holds the promise to address challenging robotic tasks such as autonomous navigation. It however requires a human supervisor to oversee the training process and send correct control commands to robots without feedback, which is always prone to error and expensive. To minimize human involvement and avoid manual labeling of data in the robotic autonomous navigation with imitation learning, this paper proposes a novel semi-supervised imitation learning solution based on a multi-sensory design. This solution includes a suboptimal sensor policy based on sensor fusion to automatically label states encountered by a robot to avoid human supervision during training. In addition, a recording policy is developed to throttle the adversarial affect of learning too much from the suboptimal sensor policy. This solution allows the robot to learn a navigation policy in a self-supervised manner. With extensive experiments in indoor environments, this solution can achieve near human performance in most of the tasks and even surpasses human performance in case of unexpected events such as hardware failures or human operation errors. To best of our knowledge, this is the first work that synthesizes sensor fusion and imitation learning to enable robotic autonomous navigation in the real world without human supervision.