Search

Model-based reinforcement learning has shown promise for improving sample efficiency and decision-making in complex environments. However, existing methods face challenges in training stability, robustness to noise, and computational efficiency. In this paper, we propose Bisimulation Metric for Model Predictive Control (BS-MPC), a novel approach that incorporates bisimulation metric loss in its objective function to directly optimize the encoder. This time-step-wise direct optimization enables the learned encoder to extract intrinsic information from the original state space while discarding irrelevant details and preventing the gradients and errors from diverging. BS-MPC improves training stability, robustness against input noise, and computational efficiency by reducing training time. We evaluate BS-MPC on both continuous control and image-based tasks from the DeepMind Control Suite, demonstrating superior performance and robustness compared to state-of-the-art baseline methods.

Offline reinforcement learning (RL) is a compelling paradigm to extend RL's practical utility by leveraging pre-collected, static datasets, thereby avoiding the limitations associated with collecting online interactions. The major difficulty in offline RL is mitigating the impact of approximation errors when encountering out-of-distribution (OOD) actions; doing so ineffectively will lead to policies that prefer OOD actions, which can lead to unexpected and potentially catastrophic results. Despite the variety of works proposed to address this issue, they tend to excessively suppress the value function in and around OOD regions, resulting in overly pessimistic value estimates. In this paper, we propose a novel framework called Strategically Conservative Q-Learning (SCQ) that distinguishes between OOD data that is easy and hard to estimate, ultimately resulting in less conservative value estimates. Our approach exploits the inherent strengths of neural networks to interpolate, while carefully navigating their limitations in extrapolation, to obtain pessimistic yet still property calibrated value estimates. Theoretical analysis also shows that the value function learned by SCQ is still conservative, but potentially much less so than that of Conservative Q-learning (CQL). Finally, extensive evaluation on the D4RL benchmark tasks shows our proposed method outperforms state-of-the-art methods. Our code is available through \url{https://github.com/purewater0901/SCQ}.

This paper develops a new nonlinear filter, called Moment-based Kalman Filter (MKF), using the exact moment propagation method. Existing state estimation methods use linearization techniques or sampling points to compute approximate values of moments. However, moment propagation of probability distributions of random variables through nonlinear process and measurement models play a key role in the development of state estimation and directly affects their performance. The proposed moment propagation procedure can compute exact moments for non-Gaussian as well as non-independent Gaussian random variables. Thus, MKF can propagate exact moments of uncertain state variables up to any desired order. MKF is derivative-free and does not require tuning parameters. Moreover, MKF has the same computation time complexity as the extended or unscented Kalman filters, i.e., EKF and UKF. The experimental evaluations show that MKF is the preferred filter in comparison to EKF and UKF and outperforms both filters in non-Gaussian noise regimes., Comment: Accepted at the IEEE Conference on Robotics and Automation (ICRA), 2023

Dynamic game arises as a powerful paradigm for multi-robot planning, for which safety constraint satisfaction is crucial. Constrained stochastic games are of particular interest, as real-world robots need to operate and satisfy constraints under uncertainty. Existing methods for solving stochastic games handle chance constraints using exponential penalties with hand-tuned weights. However, finding a suitable penalty weight is nontrivial and requires trial and error. In this paper, we propose the chance-constrained iterative linear-quadratic stochastic games (CCILQGames) algorithm. CCILQGames solves chance-constrained stochastic games using the augmented Lagrangian method. We evaluate our algorithm in three autonomous driving scenarios, including merge, intersection, and roundabout. Experimental results and Monte Carlo tests show that CCILQGames can generate safe and interactive strategies in stochastic environments.

Velocity Planning for self-driving vehicles in a complex environment is one of the most challenging tasks. It must satisfy the following three requirements: safety with regards to collisions; respect of the maximum velocity limits defined by the traffic rules; comfort of the passengers. In order to achieve these goals, the jerk and dynamic objects should be considered, however, it makes the problem as complex as a non-convex optimization problem. In this paper, we propose a linear programming (LP) based velocity planning method with jerk limit and obstacle avoidance constraints for an autonomous driving system. To confirm the efficiency of the proposed method, a comparison is made with several optimization-based approaches, and we show that our method can generate a velocity profile which satisfies the aforementioned requirements more efficiently than the compared methods. In addition, we tested our algorithm on a real vehicle at a test field to validate the effectiveness of the proposed method.

Motion planning under uncertainty is of significant importance for safety-critical systems such as autonomous vehicles. Such systems have to satisfy necessary constraints (e.g., collision avoidance) with potential uncertainties coming from either disturbed system dynamics or noisy sensor measurements. However, existing motion planning methods cannot efficiently find the robust optimal solutions under general nonlinear and non-convex settings. In this paper, we formulate such problem as chance-constrained Gaussian belief space planning and propose the constrained iterative Linear Quadratic Gaussian (CILQG) algorithm as a real-time solution. In this algorithm, we iteratively calculate a Gaussian approximation of the belief and transform the chance-constraints. We evaluate the effectiveness of our method in simulations of autonomous driving planning tasks with static and dynamic obstacles. Results show that CILQG can handle uncertainties more appropriately and has faster computation time than baseline methods., Comment: IROS 2021

We have found a radio-wave-reflection effect in rock salt for the detection of ultra-high energy neutrinos which are expected to be generated in Greisen, Zatsepin, and Kuzmin (GZK) processes in the universe. When an UHE neutrino interacts with rock salt or ice as a detection medium, a shower is generated. That shower is formed by hadronic and electromagnetic avalanche processes. The energy of the UHE neutrino shower converts to thermal energy through ionization processes. Consequently, the temperature rises along the shower produced by the UHE neutrino. The refractive index of the medium rises with temperature. The irregularity of the refractive index in the medium leads to a reflection of radio waves. This reflection effect combined with the long attenuation length of radio waves in rock salt and ice would yield a new method to detect UHE neutrinos. We measured the phase of the reflected radio wave under irradiation with an electron beam on ice and rock salt powder. The measured phase showed excellent consistence with the power reflection fraction which was measured directly. A model taking into account the temperature change explained the phase and the amplitude of the reflected wave. Therefore the reflection mechanism was confirmed. The power reflection fraction was compared with that calculated with the Fresnel equations, the ratio between the measured result and that obtained with the Fresnel equations in ice was larger than that of rock salt., Comment: 6 Pages, 5th International Workshop on Acoustic and Radio EeV Neutrino Detection Activities

Existence of GZK neutrinos (ultra high energy neutrinos) have been justified although the flux is very low. A new method is desired to use a huge mass of a detector medium to detect them. A fundamental study of radar method was carried out to measure microwave reflection from electromagnetic energy deposit by X-ray irradiation in a small rock salt sample. The reflection rate of 1x10^-6 was found at the energy deposit of 1x10^19 eV which was proportional to square of the X-ray intensity suggesting the effect to be coherent scattering. The decay time of the reflection was several seconds. This effect implies a large scale natural rock salt formation could be utilized like a bubble chamber irradiated by radio wave instead of visible light to detect GZK neutrinos., Comment: Submitted for the SUSY07 proceedings, 4 pages, LaTeX, 3 eps figures