LoKO: Low-Rank Kalman Optimizer for Online Fine-Tuning of Large Models

Training large models with millions or even billions of parameters from scratch incurs substantial computational costs. Parameter Efficient Fine-Tuning (PEFT) methods, particularly Low-Rank Adaptation (LoRA), address this challenge by adapting only a reduced number of parameters to specific tasks with gradient-based optimizers. In this paper, we cast PEFT as an optimal filtering/state estimation problem and present Low-Rank Kalman Optimizer (LoKO) to estimate the optimal trainable parameters in an online manner. We leverage the low-rank decomposition in LoRA to significantly reduce matrix sizes in Kalman iterations and further capitalize on a diagonal approximation of the covariance matrix to effectively decrease computational complexity from quadratic to linear in the number of trainable parameters. Moreover, we discovered that the initialization of the covariance matrix within the Kalman algorithm and the accurate estimation of the observation noise covariance are the keys in this formulation, and we propose robust approaches that work well across a vast range of well-established computer vision and language models. Our results show that LoKO converges with fewer iterations and yields better performance models compared to commonly used optimizers with LoRA in both image classifications and language tasks. Our study opens up the possibility of leveraging the Kalman filter as an effective optimizer for the online fine-tuning of large models.

Modular Adaptive Aerial Manipulation under Unknown Dynamic Coupling Forces

Successful aerial manipulation largely depends on how effectively a controller can tackle the coupling dynamic forces between the aerial vehicle and the manipulator. However, this control problem has remained largely unsolved as the existing control approaches either require precise knowledge of the aerial vehicle/manipulator inertial couplings, or neglect the state-dependent uncertainties especially arising during the interaction phase. This work proposes an adaptive control solution to overcome this long standing control challenge without any a priori knowledge of the coupling dynamic terms. Additionally, in contrast to the existing adaptive control solutions, the proposed control framework is modular, that is, it allows independent tuning of the adaptive gains for the vehicle position sub-dynamics, the vehicle attitude sub-dynamics, and the manipulator sub-dynamics. Stability of the closed loop under the proposed scheme is derived analytically, and real-time experiments validate the effectiveness of the proposed scheme over the state-of-the-art approaches.

Spatial-aware decision-making with ring attractors in reinforcement learning systems

This paper explores the integration of ring attractors, a mathematical model inspired by neural circuit dynamics, into the reinforcement learning (RL) action selection process. Ring attractors, as specialized brain-inspired structures that encode spatial information and uncertainty, offer a biologically plausible mechanism to improve learning speed and predictive performance. They do so by explicitly encoding the action space, facilitating the organization of neural activity, and enabling the distribution of spatial representations across the neural network in the context of deep RL. The application of ring attractors in the RL action selection process involves mapping actions to specific locations on the ring and decoding the selected action based on neural activity. We investigate the application of ring attractors by both building them as exogenous models and integrating them as part of a Deep Learning policy algorithm. Our results show a significant improvement in state-of-the-art models for the Atari 100k benchmark. Notably, our integrated approach improves the performance of state-of-the-art models by half, representing a 53\% increase over selected baselines.

DroneDiffusion: Robust Quadrotor Dynamics Learning with Diffusion Models

An inherent fragility of quadrotor systems stems from model inaccuracies and external disturbances. These factors hinder performance and compromise the stability of the system, making precise control challenging. Existing model-based approaches either make deterministic assumptions, utilize Gaussian-based representations of uncertainty, or rely on nominal models, all of which often fall short in capturing the complex, multimodal nature of real-world dynamics. This work introduces DroneDiffusion, a novel framework that leverages conditional diffusion models to learn quadrotor dynamics, formulated as a sequence generation task. DroneDiffusion achieves superior generalization to unseen, complex scenarios by capturing the temporal nature of uncertainties and mitigating error propagation. We integrate the learned dynamics with an adaptive controller for trajectory tracking with stability guarantees. Extensive experiments in both simulation and real-world flights demonstrate the robustness of the framework across a range of scenarios, including unfamiliar flight paths and varying payloads, velocities, and wind disturbances.

HOLA-Drone: Hypergraphic Open-ended Learning for Zero-Shot Multi-Drone Cooperative Pursuit

Zero-shot coordination (ZSC) is a significant challenge in multi-agent collaboration, aiming to develop agents that can coordinate with unseen partners they have not encountered before. Recent cutting-edge ZSC methods have primarily focused on two-player video games such as OverCooked!2 and Hanabi. In this paper, we extend the scope of ZSC research to the multi-drone cooperative pursuit scenario, exploring how to construct a drone agent capable of coordinating with multiple unseen partners to capture multiple evaders. We propose a novel Hypergraphic Open-ended Learning Algorithm (HOLA-Drone) that continuously adapts the learning objective based on our hypergraphic-form game modeling, aiming to improve cooperative abilities with multiple unknown drone teammates. To empirically verify the effectiveness of HOLA-Drone, we build two different unseen drone teammate pools to evaluate their performance in coordination with various unseen partners. The experimental results demonstrate that HOLA-Drone outperforms the baseline methods in coordination with unseen drone teammates. Furthermore, real-world experiments validate the feasibility of HOLA-Drone in physical systems. Videos can be found on the project homepage~\url{https://sites.google.com/view/hola-drone}.

Adaptive Advantage-Guided Policy Regularization for Offline Reinforcement Learning

In offline reinforcement learning, the challenge of out-of-distribution (OOD) is pronounced. To address this, existing methods often constrain the learned policy through policy regularization. However, these methods often suffer from the issue of unnecessary conservativeness, hampering policy improvement. This occurs due to the indiscriminate use of all actions from the behavior policy that generates the offline dataset as constraints. The problem becomes particularly noticeable when the quality of the dataset is suboptimal. Thus, we propose Adaptive Advantage-guided Policy Regularization (A2PR), obtaining high-advantage actions from an augmented behavior policy combined with VAE to guide the learned policy. A2PR can select high-advantage actions that differ from those present in the dataset, while still effectively maintaining conservatism from OOD actions. This is achieved by harnessing the VAE capacity to generate samples matching the distribution of the data points. We theoretically prove that the improvement of the behavior policy is guaranteed. Besides, it effectively mitigates value overestimation with a bounded performance gap. Empirically, we conduct a series of experiments on the D4RL benchmark, where A2PR demonstrates state-of-the-art performance. Furthermore, experimental results on additional suboptimal mixed datasets reveal that A2PR exhibits superior performance. Code is available at https://github.com/ltlhuuu/A2PR.

TTA-Nav: Test-time Adaptive Reconstruction for Point-Goal Navigation under Visual Corruptions

Robot navigation under visual corruption presents a formidable challenge. To address this, we propose a Test-time Adaptation (TTA) method, named as TTA-Nav, for point-goal navigation under visual corruptions. Our "plug-and-play" method incorporates a top-down decoder to a pre-trained navigation model. Firstly, the pre-trained navigation model gets a corrupted image and extracts features. Secondly, the top-down decoder produces the reconstruction given the high-level features extracted by the pre-trained model. Then, it feeds the reconstruction of a corrupted image back to the pre-trained model. Finally, the pre-trained model does forward pass again to output action. Despite being trained solely on clean images, the top-down decoder can reconstruct cleaner images from corrupted ones without the need for gradient-based adaptation. The pre-trained navigation model with our top-down decoder significantly enhances navigation performance across almost all visual corruptions in our benchmarks. Our method improves the success rate of point-goal navigation from the state-of-the-art result of 46% to 94% on the most severe corruption. This suggests its potential for broader application in robotic visual navigation. Project page: https://sites.google.com/view/tta-nav

Impact of Computation in Integral Reinforcement Learning for Continuous-Time Control

Integral reinforcement learning (IntRL) demands the precise computation of the utility function's integral at its policy evaluation (PEV) stage. This is achieved through quadrature rules, which are weighted sums of utility functions evaluated from state samples obtained in discrete time. Our research reveals a critical yet underexplored phenomenon: the choice of the computational method -- in this case, the quadrature rule -- can significantly impact control performance. This impact is traced back to the fact that computational errors introduced in the PEV stage can affect the policy iteration's convergence behavior, which in turn affects the learned controller. To elucidate how computation impacts control, we draw a parallel between IntRL's policy iteration and Newton's method applied to the Hamilton-Jacobi-Bellman equation. In this light, computational error in PEV manifests as an extra error term in each iteration of Newton's method, with its upper bound proportional to the computational error. Further, we demonstrate that when the utility function resides in a reproducing kernel Hilbert space (RKHS), the optimal quadrature is achievable by employing Bayesian quadrature with the RKHS-inducing kernel function. We prove that the local convergence rates for IntRL using the trapezoidal rule and Bayesian quadrature with a Mat\'ern kernel to be $O(N^{-2})$ and $O(N^{-b})$, where $N$ is the number of evenly-spaced samples and $b$ is the Mat\'ern kernel's smoothness parameter. These theoretical findings are finally validated by two canonical control tasks.

Open Ad Hoc Teamwork with Cooperative Game Theory

Ad hoc teamwork poses a challenging problem, requiring the design of an agent to collaborate with teammates without prior coordination or joint training. Open ad hoc teamwork further complicates this challenge by considering environments with a changing number of teammates, referred to as open teams. The state-of-the-art solution to this problem is graph-based policy learning (GPL), leveraging the generalizability of graph neural networks to handle an unrestricted number of agents and effectively address open teams. GPL's performance is superior to other methods, but its joint Q-value representation presents challenges for interpretation, hindering further development of this research line and applicability. In this paper, we establish a new theory to give an interpretation for the joint Q-value representation employed in GPL, from the perspective of cooperative game theory. Building on our theory, we propose a novel algorithm based on GPL framework, to complement the critical features that facilitate learning, but overlooked in GPL. Through experiments, we demonstrate the correctness of our theory by comparing the performance of the resulting algorithm with GPL in dynamic team compositions.

Aligning Individual and Collective Objectives in Multi-Agent Cooperation

In the field of multi-agent learning, the challenge of mixed-motive cooperation is pronounced, given the inherent contradictions between individual and collective goals. Current research in this domain primarily focuses on incorporating domain knowledge into rewards or introducing additional mechanisms to foster cooperation. However, many of these methods suffer from the drawbacks of manual design costs and the lack of a theoretical grounding convergence procedure to the solution. To address this gap, we approach the mixed-motive game by modeling it as a differentiable game to study learning dynamics. We introduce a novel optimization method named Altruistic Gradient Adjustment (AgA) that employs gradient adjustments to novelly align individual and collective objectives. Furthermore, we provide theoretical proof that the selection of an appropriate alignment weight in AgA can accelerate convergence towards the desired solutions while effectively avoiding the undesired ones. The visualization of learning dynamics effectively demonstrates that AgA successfully achieves alignment between individual and collective objectives. Additionally, through evaluations conducted on established mixed-motive benchmarks such as the public good game, Cleanup, Harvest, and our modified mixed-motive SMAC environment, we validate AgA's capability to facilitate altruistic and fair collaboration.