Recently, robust reinforcement learning (RL) methods designed to handle adversarial input observations have received significant attention, motivated by RL’s inherent vulnerabilities. While existing approaches have demonstrated reasonable success, addressing worst-case scenarios over long time horizons requires both minimizing the agent’s cumulative rewards for adversaries and training agents to counteract them through alternating learning. However, this process introduces mutual dependencies between the agent and the adversary, making interactions with the environment inefficient and hindering the development of off-policy methods. In this work, we propose a novel off-policy method that eliminates the need for additional environmental interactions by reformulating adversarial learning as a soft-constrained optimization problem. Our approach is theoretically supported by the symmetric property of policy evaluation between the agent and the adversary. The implementation is available at https://github.com/nakanakakosuke/VALT_SAC.

This paper proposes a trajectory tracking control law for a mobile robot with two front differential wheels and a tail wheel. The dynamics is given by mimicking Ackerman steering model for the dynamics of position and orientation, associated with the actuator dynamics of the tail wheel's angle modeled by a first-order response with respect to the robot's angular velocity. First we develop a nominal trajectory tracking control law to track a given desired trajectory by applying differential-flatness property of the unicycle model and backstepping approach to handle the actuator dynamics. The effectiveness of the trajectory tracking is demonstrated by conducting hardware robot experiment after performing system identification, which illustrates the superior performance over a benchmark method. The design is also extended to an adaptive tracking control under parameter uncertainty in the tail wheel dynamics through introducing the adaptation law of the parameters, and the performance is demonstrated in numerical simulation.

Traffic warning systems have attracted attention for reducing traffic accidents caused by motorcycles. The authors previously developed a highly portable system that uses an in-vehicle web camera and a smartphone to detect motorcycles approaching at high speed based on rear-camera input images and notifies the driver of danger. In this study, we propose a system that combines deep-learning-based object detection and heuristics to detect encroaching vehicles from images taken by a front-facing in-vehicle camera. First, we consider the shape of the bounding box of the detected vehicle to identify the encroaching vehicle. Since the encroaching vehicle is turned sideways when entering the main road, we identify side-facing vehicles by calculating the aspect ratio of the bounding box. For motorcycles, the criterion for distinguishing between parked motorcycles and encroaching motorcycles is the presence or absence of a rider. To distinguish between parked vehicles and moving vehicles, the values of the background pixels around the detected vehicle are compared with the values of pixels around the corresponding vehicle in the previous frame and the difference is used to determine whether the vehicle is moving. The results show that the above heuristics improve the accuracy of detecting encroaching vehicles.

Open World Object Detection (OWOD) is a challenging computer vision task that extends standard object detection by (1) detecting and classifying unknown objects without supervision, and (2) incrementally learning new object classes without forgetting previously learned ones. The absence of ground truths for unknown objects makes OWOD tasks particularly challenging. Many methods have addressed this by using pseudo-labels for unknown objects. The recently proposed Probabilistic Objectness transformer-based open-world detector (PROB) is a state-of-the-art model that does not require pseudo-labels for unknown objects, as it predicts probabilistic objectness. However, this method faces issues with learning conflicts between objectness and class predictions. To address this issue and further enhance performance, we propose a novel model, Decoupled PROB. Decoupled PROB introduces Early Termination of Objectness Prediction (ETOP) to stop objectness predictions at appropriate layers in the decoder, resolving the learning conflicts between class and objectness predictions in PROB. Additionally, we introduce Task-Decoupled Query Initialization (TDQI), which efficiently extracts features of known and unknown objects, thereby improving performance. TDQI is a query initialization method that combines query selection and learnable queries, and it is a module that can be easily integrated into existing DETR-based OWOD models. Extensive experiments on OWOD benchmarks demonstrate that Decoupled PROB surpasses all existing methods across several metrics, significantly improving performance.

For safety-critical robotics applications such as autonomous driving, it is important to detect all required objects accurately in real-time. Motion segmentation offers a solution by identifying dynamic objects from the scene in a class-agnostic manner. Recently, various motion segmentation models have been proposed, most of which jointly use subnetworks to estimate Depth, Pose, Optical Flow, and Scene Flow. As a result, the overall computational cost of the model increases, hindering real-time performance.In this paper, we propose a novel cost-volume-based motion feature representation, Channel-wise Motion Features. By extracting depth features of each instance in the feature map and capturing the scene’s 3D motion information, it offers enhanced efficiency. The only subnetwork used to build Channel-wise Motion Features is the Pose Network, and no others are required. Our method not only achieves about 4 times the FPS of state-of-the-art models in the KITTI Dataset and Cityscapes of the VCAS-Motion Dataset, but also demonstrates equivalent accuracy while reducing the parameters to about 25%.

To reduce motorcycle accidents, recent efforts have focused on traffic warning systems. In this study, we developed a highly portable system that utilizes smartphones and web cameras to detect motorcycles approaching at high speeds and to notify drivers of potential dangers. Specifically, we have implemented a feature to detect hazardous scenarios where motorcycles change lanes from behind and overtake on the right. Accurate lane estimation is crucial for precisely assessing situations where motorcycles change lanes. While lane estimation methods based on recognizing white road markings exist, they may not be effective on roads with unclear or faded white lines, which are common in developing countries. An alternative approach involves estimating lanes from fixed camera vehicle trajectories, but this is limited to locations where fixed cameras are installed. Moreover, deep learning methods for lane detection face computational challenges on mobile systems, especially when analyzing both lane positions and vehicle behavior. Here we propose a method to detect vehicles and estimate lanes based on their trajectories from videos captured by in-vehicle cameras in developing countries. The effectiveness of our proposed method was demonstrated through real-world road tests in Indonesia.

Vision-based object detection and tracking play a critical role in the field of autonomous driving as they enable the ego-car to identify potential dangers and detect objects that emerge suddenly. However, the current frequency of out-of-state vision-based object detection and tracking perception is limited to approximately 10-20Hz, which is insufficient for the autonomous driving system to promptly respond to events such as the sudden appearance of objects in close proximity or high-speed scenarios. In contrast, animals possess the ability to promptly avoid dangers despite lacking precise knowledge of an object’s exact distance, owing to their eyes’ high flicker fusion rates. Consequently, this paper introduces a novel vision-based hybrid object tracking technique. The proposed approach incorporates two distinct perception systems: a multi-camera perception system (MCP) and a high-frequency perception system (HFP). These systems operate independently to detect objects, with the MCP offering higher accuracy albeit lower update rates, and the HFP providing lower accuracy but higher update rates. Experimental results demonstrate that our method surpasses traditional perception approaches by achieving faster detection of real-time pop-out motorbikes across various ego-car speeds. Moreover, when both MCP and HFP information are available, our proposed method effectively combines the strengths of both systems to estimate the object’s status.

This paper deals with a kinematic analysis of locomotive cooperation necessary for two mobile robots to carry a long bar along a general wavy road. The robots are assumed to be driven by a crawler mechanism and to have a single link arm which moves in a prismatic way. The robots have a task to carry a long bar keeping its height and length constant and its posture horizontal. The analysis of kinematic conditions done for a sinusoidal wavy road is expanded for a general wavy road and the authors find that there exist two kinds of cooperation mode: one is to change the direction of locomotion and the other is to alternate the robot role from "master" to "servant" and vis-vasa.