Pretraining in Actor-Critic Reinforcement Learning for Robot Motion Control

The pretraining-finetuning paradigm has facilitated numerous transformative advancements in artificial intelligence research in recent years. However, in the domain of reinforcement learning (RL) for robot motion control, individual skills are often learned from scratch despite the high likelihood that some generalizable knowledge is shared across all task-specific policies belonging to a single robot embodiment. This work aims to define a paradigm for pretraining neural network models that encapsulate such knowledge and can subsequently serve as a basis for warm-starting the RL process in classic actor-critic algorithms, such as Proximal Policy Optimization (PPO). We begin with a task-agnostic exploration-based data collection algorithm to gather diverse, dynamic transition data, which is then used to train a Proprioceptive Inverse Dynamics Model (PIDM) through supervised learning. The pretrained weights are loaded into both the actor and critic networks to warm-start the policy optimization of actual tasks. We systematically validated our proposed method on seven distinct robot motion control tasks, showing significant benefits to this initialization strategy. Our proposed approach on average improves sample efficiency by 40.1% and task performance by 7.5%, compared to random initialization. We further present key ablation studies and empirical analyses that shed light on the mechanisms behind the effectiveness of our method.

What Matters in RL-Based Methods for Object-Goal Navigation? An Empirical Study and A Unified Framework

Object-Goal Navigation (ObjectNav) is a critical component toward deploying mobile robots in everyday, uncontrolled environments such as homes, schools, and workplaces. In this context, a robot must locate target objects in previously unseen environments using only its onboard perception. Success requires the integration of semantic understanding, spatial reasoning, and long-horizon planning, which is a combination that remains extremely challenging. While reinforcement learning (RL) has become the dominant paradigm, progress has spanned a wide range of design choices, yet the field still lacks a unifying analysis to determine which components truly drive performance. In this work, we conduct a large-scale empirical study of modular RL-based ObjectNav systems, decomposing them into three key components: perception, policy, and test-time enhancement. Through extensive controlled experiments, we isolate the contribution of each and uncover clear trends: perception quality and test-time strategies are decisive drivers of performance, whereas policy improvements with current methods yield only marginal gains. Building on these insights, we propose practical design guidelines and demonstrate an enhanced modular system that surpasses State-of-the-Art (SotA) methods by 6.6% on SPL and by a 2.7% success rate. We also introduce a human baseline under identical conditions, where experts achieve an average 98% success, underscoring the gap between RL agents and human-level navigation. Our study not only sets the SotA performance but also provides principled guidance for future ObjectNav development and evaluation.

ExT: Towards Scalable Autonomous Excavation via Large-Scale Multi-Task Pretraining and Fine-Tuning

Scaling up the deployment of autonomous excavators is of great economic and societal importance. Yet it remains a challenging problem, as effective systems must robustly handle unseen worksite conditions and new hardware configurations. Current state-of-the-art approaches rely on highly engineered, task-specific controllers, which require extensive manual tuning for each new scenario. In contrast, recent advances in large-scale pretrained models have shown remarkable adaptability across tasks and embodiments in domains such as manipulation and navigation, but their applicability to heavy construction machinery remains largely unexplored. In this work, we introduce ExT, a unified open-source framework for large-scale demonstration collection, pretraining, and fine-tuning of multitask excavation policies. ExT policies are first trained on large-scale demonstrations collected from a mix of experts, then fine-tuned either with supervised fine-tuning (SFT) or reinforcement learning fine-tuning (RLFT) to specialize to new tasks or operating conditions. Through both simulation and real-world experiments, we show that pretrained ExT policies can execute complete excavation cycles with centimeter-level accuracy, successfully transferring from simulation to real machine with performance comparable to specialized single-task controllers. Furthermore, in simulation, we demonstrate that ExT's fine-tuning pipelines allow rapid adaptation to new tasks, out-of-distribution conditions, and machine configurations, while maintaining strong performance on previously learned tasks. These results highlight the potential of ExT to serve as a foundation for scalable and generalizable autonomous excavation.

Collaborative Loco-Manipulation for Pick-and-Place Tasks with Dynamic Reward Curriculum

We present a hierarchical RL pipeline for training one-armed legged robots to perform pick-and-place (P&P) tasks end-to-end -- from approaching the payload to releasing it at a target area -- in both single-robot and cooperative dual-robot settings. We introduce a novel dynamic reward curriculum that enables a single policy to efficiently learn long-horizon P&P operations by progressively guiding the agents through payload-centered sub-objectives. Compared to state-of-the-art approaches for long-horizon RL tasks, our method improves training efficiency by 55% and reduces execution time by 18.6% in simulation experiments. In the dual-robot case, we show that our policy enables each robot to attend to different components of its observation space at distinct task stages, promoting effective coordination via autonomous attention shifts. We validate our method through real-world experiments using ANYmal D platforms in both single- and dual-robot scenarios. To our knowledge, this is the first RL pipeline that tackles the full scope of collaborative P&P with two legged manipulators.

Divide, Discover, Deploy: Factorized Skill Learning with Symmetry and Style Priors

Unsupervised Skill Discovery (USD) allows agents to autonomously learn diverse behaviors without task-specific rewards. While recent USD methods have shown promise, their application to real-world robotics remains underexplored. In this paper, we propose a modular USD framework to address the challenges in the safety, interpretability, and deployability of the learned skills. Our approach employs user-defined factorization of the state space to learn disentangled skill representations. It assigns different skill discovery algorithms to each factor based on the desired intrinsic reward function. To encourage structured morphology-aware skills, we introduce symmetry-based inductive biases tailored to individual factors. We also incorporate a style factor and regularization penalties to promote safe and robust behaviors. We evaluate our framework in simulation using a quadrupedal robot and demonstrate zero-shot transfer of the learned skills to real hardware. Our results show that factorization and symmetry lead to the discovery of structured human-interpretable behaviors, while the style factor and penalties enhance safety and diversity. Additionally, we show that the learned skills can be used for downstream tasks and perform on par with oracle policies trained with hand-crafted rewards.

GraspQP: Differentiable Optimization of Force Closure for Diverse and Robust Dexterous Grasping

Dexterous robotic hands enable versatile interactions due to the flexibility and adaptability of multi-fingered designs, allowing for a wide range of task-specific grasp configurations in diverse environments. However, to fully exploit the capabilities of dexterous hands, access to diverse and high-quality grasp data is essential -- whether for developing grasp prediction models from point clouds, training manipulation policies, or supporting high-level task planning with broader action options. Existing approaches for dataset generation typically rely on sampling-based algorithms or simplified force-closure analysis, which tend to converge to power grasps and often exhibit limited diversity. In this work, we propose a method to synthesize large-scale, diverse, and physically feasible grasps that extend beyond simple power grasps to include refined manipulations, such as pinches and tri-finger precision grasps. We introduce a rigorous, differentiable energy formulation of force closure, implicitly defined through a Quadratic Program (QP). Additionally, we present an adjusted optimization method (MALA*) that improves performance by dynamically rejecting gradient steps based on the distribution of energy values across all samples. We extensively evaluate our approach and demonstrate significant improvements in both grasp diversity and the stability of final grasp predictions. Finally, we provide a new, large-scale grasp dataset for 5,700 objects from DexGraspNet, comprising five different grippers and three distinct grasp types. Dataset and Code:https://graspqp.github.io/

Large Scale Robotic Material Handling: Learning, Planning, and Control

Bulk material handling involves the efficient and precise moving of large quantities of materials, a core operation in many industries, including cargo ship unloading, waste sorting, construction, and demolition. These repetitive, labor-intensive, and safety-critical operations are typically performed using large hydraulic material handlers equipped with underactuated grippers. In this work, we present a comprehensive framework for the autonomous execution of large-scale material handling tasks. The system integrates specialized modules for environment perception, pile attack point selection, path planning, and motion control. The main contributions of this work are two reinforcement learning-based modules: an attack point planner that selects optimal grasping locations on the material pile to maximize removal efficiency and minimize the number of scoops, and a robust trajectory following controller that addresses the precision and safety challenges associated with underactuated grippers in movement, while utilizing their free-swinging nature to release material through dynamic throwing. We validate our framework through real-world experiments on a 40 t material handler in a representative worksite, focusing on two key tasks: high-throughput bulk pile management and high-precision truck loading. Comparative evaluations against human operators demonstrate the system's effectiveness in terms of precision, repeatability, and operational safety. To the best of our knowledge, this is the first complete automation of material handling tasks on a full scale.

Experience is the Best Teacher: Grounding VLMs for Robotics through Self-Generated Memory

Vision-language models (VLMs) have been widely adopted in robotics to enable autonomous planning. However, grounding VLMs, originally trained on internet data, to diverse real-world robots remains a challenge. This paper presents ExpTeach, a framework that grounds VLMs to physical robots by building a self-generated memory of real-world experiences. In ExpTeach, the VLM autonomously plans actions, verifies outcomes, reflects on failures, and adapts robot behaviors in a closed loop. The self-generated experiences during this process are then summarized into a long-term memory, enabling retrieval of learned knowledge to guide future tasks via retrieval-augmented generation (RAG). Additionally, ExpTeach enhances the spatial understanding of VLMs with an on-demand image annotation module. In experiments, we show that reflection improves success rates from 36% to 84% on four challenging robotic tasks and observe the emergence of intelligent object interactions, including creative tool use. Across extensive tests on 12 real-world scenarios (including eight unseen ones), we find that grounding with long-term memory boosts single-trial success rates from 22% to 80%, demonstrating the effectiveness and generalizability of ExpTeach.

Attention-Based Map Encoding for Learning Generalized Legged Locomotion

Dynamic locomotion of legged robots is a critical yet challenging topic in expanding the operational range of mobile robots. It requires precise planning when possible footholds are sparse, robustness against uncertainties and disturbances, and generalizability across diverse terrains. While traditional model-based controllers excel at planning on complex terrains, they struggle with real-world uncertainties. Learning-based controllers offer robustness to such uncertainties but often lack precision on terrains with sparse steppable areas. Hybrid methods achieve enhanced robustness on sparse terrains by combining both methods but are computationally demanding and constrained by the inherent limitations of model-based planners. To achieve generalized legged locomotion on diverse terrains while preserving the robustness of learning-based controllers, this paper proposes to learn an attention-based map encoding conditioned on robot proprioception, which is trained as part of the end-to-end controller using reinforcement learning. We show that the network learns to focus on steppable areas for future footholds when the robot dynamically navigates diverse and challenging terrains. We synthesize behaviors that exhibit robustness against uncertainties while enabling precise and agile traversal of sparse terrains. Additionally, our method offers a way to interpret the topographical perception of a neural network. We have trained two controllers for a 12-DoF quadrupedal robot and a 23-DoF humanoid robot respectively and tested the resulting controllers in the real world under various challenging indoor and outdoor scenarios, including ones unseen during training.

Improving Long-Range Navigation with Spatially-Enhanced Recurrent Memory via End-to-End Reinforcement Learning

Recent advancements in robot navigation, especially with end-to-end learning approaches like reinforcement learning (RL), have shown remarkable efficiency and effectiveness. Yet, successful navigation still relies on two key capabilities: mapping and planning, whether explicit or implicit. Classical approaches use explicit mapping pipelines to register ego-centric observations into a coherent map frame for the planner. In contrast, end-to-end learning achieves this implicitly, often through recurrent neural networks (RNNs) that fuse current and past observations into a latent space for planning. While architectures such as LSTM and GRU capture temporal dependencies, our findings reveal a key limitation: their inability to perform effective spatial memorization. This skill is essential for transforming and integrating sequential observations from varying perspectives to build spatial representations that support downstream planning. To address this, we propose Spatially-Enhanced Recurrent Units (SRUs), a simple yet effective modification to existing RNNs, designed to enhance spatial memorization capabilities. We introduce an attention-based architecture with SRUs, enabling long-range navigation using a single forward-facing stereo camera. Regularization techniques are employed to ensure robust end-to-end recurrent training via RL. Experimental results show our approach improves long-range navigation by 23.5% compared to existing RNNs. Furthermore, with SRU memory, our method outperforms the RL baseline with explicit mapping and memory modules, achieving a 29.6% improvement in diverse environments requiring long-horizon mapping and memorization. Finally, we address the sim-to-real gap by leveraging large-scale pretraining on synthetic depth data, enabling zero-shot transfer to diverse and complex real-world environments.