Recent advances in LLM agents enable systems that autonomously refine workflows, accumulate reusable skills, self-train their underlying models, and maintain persistent memory. However, we show that such self-evolution is often non-monotonic: adapting to new task distributions can progressively degrade previously acquired capabilities across all major evolution channels.
We identify this phenomenon as \emph{capability erosion under self-evolution} and show that it consistently emerges across workflow, skill, model, and memory evolution. To mitigate this issue, we propose \emph{Capability-Preserving Evolution} (CPE), a general stabilization principle that constrains destructive capability drift during continual adaptation. Across all four evolution dimensions, CPE consistently improves retained capability stability while preserving adaptation performance. For example, in workflow evolution, CPE improves retained simple-task performance from 41.8\% to 52.8\% under GPT-5.1 optimization while simultaneously achieving stronger complex-task adaptation.
Our findings suggest that stable long-horizon self-evolving agents require not only acquiring new capabilities, but also explicitly preserving previously learned ones during continual adaptation.
KU Leuven（鲁汶大学）；Pontificia Universidad Católica de Chile（智利天主教大学）；University of Pisa（比萨大学）
Lifelong language learning seeks to have models continuously learn multiple tasks in a sequential order without suffering from catastrophic forgetting. State-of-the-art approaches rely on sparse experience replay as the primary approach to prevent forgetting. Experience replay usually adopts sampling methods for the memory population; however, the effect of the chosen sampling strategy on model performance has not yet been studied. In this paper, we investigate how relevant the selective memory population is in the lifelong learning process of text classification and question-answering tasks. …
Continual learning (CL) aims to enable information systems to learn from a continuous data stream across time. However, it is difficult for existing deep learning architectures to learn a new task without largely forgetting previously acquired knowledge. Furthermore, CL is particularly challenging for language learning, as natural language is ambiguous: it is discrete, compositional, and its meaning is context-dependent. In this work, we look at the problem of CL through the lens of various NLP tasks. Our survey discusses major challenges in CL and current methods applied in neural network mod…
State Key Laboratory of Robotics and Intelligent Systems（机器人与智能系统国家重点实验室）；Shenyang Institute of Automation（沈阳自动化研究所）；Chinese Academy of Sciences（中国科学院）；University of Chinese Academy of Sciences（中国科学院大学）；Mohamed bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）
University of Pennsylvania（宾夕法尼亚大学）；University of Western Ontario（西安大略大学）
Policy gradient methods have shown success in learning control policies for high-dimensional dynamical systems. Their biggest downside is the amount of exploration they require before yielding high-performing policies. In a lifelong learning setting, in which an agent is faced with multiple consecutive tasks over its lifetime, reusing information from previously seen tasks can substantially accelerate the learning of new tasks. We provide a novel method for lifelong policy gradient learning that trains lifelong function approximators directly via policy gradients, allowing the agent to benefit…
University of Hamburg（汉堡大学）；University of Malaya（马来亚大学）
Lifelong learning is a long-standing aim for artificial agents that act in dynamic environments, in which an agent needs to accumulate knowledge incrementally without forgetting previously learned representations. We investigate methods for learning from data produced by event cameras and compare techniques to mitigate forgetting while learning incrementally. We propose a model that is composed of both, feature extraction and continuous learning. Furthermore, we introduce a habituation-based method to mitigate forgetting. Our experimental results show that the combination of different techniqu…
Fudan University（复旦大学）；Xiamen University（厦门大学）；The Pennsylvania State University（宾夕法尼亚州立大学）；Chinese Academy of Sciences（中国科学院）
Crowd counting has important applications in public safety and pandemic control. A robust and practical crowd counting system has to be capable of continuously learning with the new incoming domain data in real-world scenarios instead of fitting one domain only. Off-the-shelf methods have some drawbacks when handling multiple domains: (1) the models will achieve limited performance (even drop dramatically) among old domains after training images from new domains due to the discrepancies of intrinsic data distributions from various domains, which is called catastrophic forgetting; (2) the well-…
Universität Hamburg（汉堡大学）；Okinawa Institute of Science and Technology（冲绳科学技术大学院大学）
Artificial autonomous agents and robots interacting in complex environments are required to continually acquire and fine-tune knowledge over sustained periods of time. The ability to learn from continuous streams of information is referred to as lifelong learning and represents a long-standing challenge for neural network models due to catastrophic forgetting. Computational models of lifelong learning typically alleviate catastrophic forgetting in experimental scenarios with given datasets of static images and limited complexity, thereby differing significantly from the conditions artificial a…
Suzhou University of Science and Technology（苏州科技大学）；Tianjin University（天津大学）
The Lifelong Multi-Label (LML) image recognition builds an online class-incremental classifier in a sequential multi-label image recognition data stream. The key challenges of LML image recognition are the construction of label relationships on Partial Labels of training data and the Catastrophic Forgetting on old classes, resulting in poor generalization. To solve the problems, the study proposes an Augmented Graph Convolutional Network (AGCN) model that can construct the label relationships across the sequential recognition tasks and sustain the catastrophic forgetting. First, we build an Au…
Mila（米拉研究所）；McGill University（麦吉尔大学）；Université de Montréal（蒙特利尔大学）
To achieve general artificial intelligence, reinforcement learning (RL) agents should learn not only to optimize returns for one specific task but also to constantly build more complex skills and scaffold their knowledge about the world, without forgetting what has already been learned. In this paper, we discuss the desired characteristics of environments that can support the training and evaluation of lifelong reinforcement learning agents, review existing environments from this perspective, and propose recommendations for devising suitable environments in the future.
This paper introduces a new lifelong learning solution where a single model is trained for a sequence of tasks. The main challenge that vision systems face in this context is catastrophic forgetting: as they tend to adapt to the most recently seen task, they lose performance on the tasks that were learned previously. Our method aims at preserving the knowledge of the previous tasks while learning a new one by using autoencoders. For each task, an under-complete autoencoder is learned, capturing the features that are crucial for its achievement. When a new task is presented to the system, we pr…
One approach to meet the challenges of deep lifelong reinforcement learning (LRL) is careful management of the agent's learning experiences, to learn (without forgetting) and build internal meta-models (of the tasks, environments, agents, and world). Generative replay (GR) is a biologically inspired replay mechanism that augments learning experiences with self-labelled examples drawn from an internal generative model that is updated over time. We present a version of GR for LRL that satisfies two desiderata: (a) Introspective density modelling of the latent representations of policies learned …
Task-agnostic Lifelong Robot Learning with Retrieval-based Weighted Local Adaptation
College of Design and Engineering National University of Singapore（设计与工程学院新加坡国立大学）；Jacobs School of Engineering University of California San Diego（工程学院加州大学圣地亚哥分校）；Massachusetts General Hospital Harvard University（麻省总医院哈佛大学）；Faculty of Electrical Engineering, Mathematics and Computer Science Delft University of Technology（电子工程、数学和计算机科学学院代尔夫特技术大学）；Faculty of Mechanical Engineering Delft University of Technology（机械工程学院代尔夫特技术大学）
SRI International（斯坦福国际研究院）；American University（美利坚大学）；Rochester Institute of Technology（罗切斯特理工学院）；Georgia Institute of Technology（佐治亚理工学院）
As Artificial and Robotic Systems are increasingly deployed and relied upon for real-world applications, it is important that they exhibit the ability to continually learn and adapt in dynamically-changing environments, becoming Lifelong Learning Machines. Continual/lifelong learning (LL) involves minimizing catastrophic forgetting of old tasks while maximizing a model's capability to learn new tasks. This paper addresses the challenging lifelong reinforcement learning (L2RL) setting. Pushing the state-of-the-art forward in L2RL and making L2RL useful for practical applications requires more t…
