Mitigating Suboptimality of Deterministic Policy Gradients in Complex Q-functions

• URL: http://arxiv.org/abs/2410.11833v2
• Date: Fri, 10 Oct 2025 08:17:12 GMT
• Title: Mitigating Suboptimality of Deterministic Policy Gradients in Complex Q-functions
• Authors: Ayush Jain, Norio Kosaka, Xinhu Li, Kyung-Min Kim, Erdem Bıyık, Joseph J. Lim,
• Abstract summary: We introduce SAVO, an actor architecture that generates multiple action proposals and selects the one with the highest Q-value.<n>We evaluate tasks such as restricted locomotion, dexterous manipulation, and large discrete-action space recommender systems.
• Score: 11.572333300040619
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: In reinforcement learning, off-policy actor-critic methods like DDPG and TD3 use deterministic policy gradients: the Q-function is learned from environment data, while the actor maximizes it via gradient ascent. We observe that in complex tasks such as dexterous manipulation and restricted locomotion with mobility constraints, the Q-function exhibits many local optima, making gradient ascent prone to getting stuck. To address this, we introduce SAVO, an actor architecture that (i) generates multiple action proposals and selects the one with the highest Q-value, and (ii) approximates the Q-function repeatedly by truncating poor local optima to guide gradient ascent more effectively. We evaluate tasks such as restricted locomotion, dexterous manipulation, and large discrete-action space recommender systems and show that our actor finds optimal actions more frequently and outperforms alternate actor architectures.

Related papers

• Push Smarter, Not Harder: Hierarchical RL-Diffusion Policy for Efficient Nonprehensile Manipulation [8.7216199131049]
  HeRD is a hierarchical reinforcement learning-diffusion policy that decomposes pushing tasks into two levels: high-level goal selection and low-level trajectory generation.<n>We employ a high-level reinforcement learning agent to select intermediate spatial goals, and a low-level goal-conditioned diffusion model to generate feasible, efficient trajectories to reach them.<n>Our results suggest that hierarchical control with generative low-level planning is a promising direction for scalable, goal-directed nonprehensile manipulation.
  arXiv  Detail & Related papers  (2025-12-10T21:40:22Z)
• Actor-Free Continuous Control via Structurally Maximizable Q-Functions [3.7193386971098406]
  We propose a purely value-based framework for continuous control that revisits structural of Q-functions.<n>We evaluate the proposed actor-free Q-learning approach on a range of standard simulation tasks.
  arXiv  Detail & Related papers  (2025-10-21T17:24:27Z)
• Q-STAC: Q-Guided Stein Variational Model Predictive Actor-Critic [12.837649598521102]
  This paper introduces the Q-guided STein variational model predictive Actor-Critic (Q-STAC) framework for continuous control tasks.<n>Our method optimize control sequences directly using learned Q-values as objectives, eliminating the need for explicit cost function design.<n>Experiments on 2D navigation and robotic manipulation tasks demonstrate that Q-STAC achieves superior sample efficiency, robustness, and optimality compared to state-of-the-art algorithms.
  arXiv  Detail & Related papers  (2025-07-09T07:53:53Z)
• Q-function Decomposition with Intervention Semantics with Factored Action Spaces [51.01244229483353]
  We consider Q-functions defined over a lower dimensional projected subspace of the original action space, and study the condition for the unbiasedness of decomposed Q-functions.<n>This leads to a general scheme which we call action decomposed reinforcement learning that uses the projected Q-functions to approximate the Q-function in standard model-free reinforcement learning algorithms.
  arXiv  Detail & Related papers  (2025-04-30T05:26:51Z)
• Digi-Q: Learning Q-Value Functions for Training Device-Control Agents [73.60512136881279]
  Digi-Q trains VLM-based action-value Q-functions which are then used to extract the agent policy. Digi-Q outperforms several prior methods on user-scale device control tasks in Android-in-the-Wild.
  arXiv  Detail & Related papers  (2025-02-13T18:55:14Z)
• Extremum-Seeking Action Selection for Accelerating Policy Optimization [18.162794442835413]
  Reinforcement learning for control over continuous spaces typically uses high-entropy policies, such as Gaussian distributions, for local exploration and estimating policy to optimize performance. We propose to improve action selection in this model-free RL setting by introducing additional adaptive control steps based on Extremum-Seeking Control (ESC) Our methods can be easily added in standard policy optimization to improve learning efficiency, which we demonstrate in various control learning environments.
  arXiv  Detail & Related papers  (2024-04-02T02:39:17Z)
• Hierarchical Diffusion Policy for Kinematics-Aware Multi-Task Robotic Manipulation [16.924613089429627]
  HDP factorises a manipulation policy into a hierarchical structure. We present a novel kinematics-aware goal-conditioned control agent. Empirically, we show that HDP achieves a significantly higher success rate than the state-of-the-art methods in both simulation and real-world.
  arXiv  Detail & Related papers  (2024-03-06T17:50:26Z)
• OIL-AD: An Anomaly Detection Framework for Sequential Decision Sequences [16.828732283348817]
  We propose an unsupervised method named Offline Learning based Anomaly Detection (OIL-AD) OIL-AD detects anomalies in decision-making sequences using two extracted behaviour features: action optimality and sequential association. Our experiments show that OIL-AD can achieve outstanding online anomaly detection performance with up to 34.8% improvement in F1 score over comparable baselines.
  arXiv  Detail & Related papers  (2024-02-07T04:06:53Z)
• Action-Quantized Offline Reinforcement Learning for Robotic Skill Learning [68.16998247593209]
  offline reinforcement learning (RL) paradigm provides recipe to convert static behavior datasets into policies that can perform better than the policy that collected the data. In this paper, we propose an adaptive scheme for action quantization. We show that several state-of-the-art offline RL methods such as IQL, CQL, and BRAC improve in performance on benchmarks when combined with our proposed discretization scheme.
  arXiv  Detail & Related papers  (2023-10-18T06:07:10Z)
• Solving Continuous Control via Q-learning [54.05120662838286]
  We show that a simple modification of deep Q-learning largely alleviates issues with actor-critic methods. By combining bang-bang action discretization with value decomposition, framing single-agent control as cooperative multi-agent reinforcement learning (MARL), this simple critic-only approach matches performance of state-of-the-art continuous actor-critic methods.
  arXiv  Detail & Related papers  (2022-10-22T22:55:50Z)
• Goal-Conditioned Q-Learning as Knowledge Distillation [136.79415677706612]
  We explore a connection between off-policy reinforcement learning in goal-conditioned settings and knowledge distillation. We empirically show that this can improve the performance of goal-conditioned off-policy reinforcement learning when the space of goals is high-dimensional. We also show that this technique can be adapted to allow for efficient learning in the case of multiple simultaneous sparse goals.
  arXiv  Detail & Related papers  (2022-08-28T22:01:10Z)
• Accelerated Policy Learning with Parallel Differentiable Simulation [59.665651562534755]
  We present a differentiable simulator and a new policy learning algorithm (SHAC) Our algorithm alleviates problems with local minima through a smooth critic function. We show substantial improvements in sample efficiency and wall-clock time over state-of-the-art RL and differentiable simulation-based algorithms.
  arXiv  Detail & Related papers  (2022-04-14T17:46:26Z)
• Learning to Walk Autonomously via Reset-Free Quality-Diversity [73.08073762433376]
  Quality-Diversity algorithms can discover large and complex behavioural repertoires consisting of both diverse and high-performing skills. Existing QD algorithms need large numbers of evaluations as well as episodic resets, which require manual human supervision and interventions. This paper proposes Reset-Free Quality-Diversity optimization (RF-QD) as a step towards autonomous learning for robotics in open-ended environments.
  arXiv  Detail & Related papers  (2022-04-07T14:07:51Z)
• Offline Reinforcement Learning with Implicit Q-Learning [85.62618088890787]
  Current offline reinforcement learning methods need to query the value of unseen actions during training to improve the policy. We propose an offline RL method that never needs to evaluate actions outside of the dataset. This method enables the learned policy to improve substantially over the best behavior in the data through generalization.
  arXiv  Detail & Related papers  (2021-10-12T17:05:05Z)
• IQ-Learn: Inverse soft-Q Learning for Imitation [95.06031307730245]
  imitation learning from a small amount of expert data can be challenging in high-dimensional environments with complex dynamics. Behavioral cloning is a simple method that is widely used due to its simplicity of implementation and stable convergence. We introduce a method for dynamics-aware IL which avoids adversarial training by learning a single Q-function.
  arXiv  Detail & Related papers  (2021-06-23T03:43:10Z)
• Recomposing the Reinforcement Learning Building Blocks with Hypernetworks [19.523737925041278]
  We show that a primary network determines the weights of a conditional dynamic network. This approach improves the gradient approximation and reduces the learning step variance. We demonstrate a consistent improvement across different locomotion tasks and different algorithms both in RL (TD3 and SAC) and in Meta-RL (MAML and PEARL)
  arXiv  Detail & Related papers  (2021-06-12T19:43:12Z)
• GRAC: Self-Guided and Self-Regularized Actor-Critic [24.268453994605512]
  We propose a self-regularized TD-learning method to address divergence without requiring a target network. We also propose a self-guided policy improvement method by combining policy-gradient with zero-order optimization. This makes learning more robust to local noise in the Q function approximation and guides the updates of our actor network. We evaluate GRAC on the suite of OpenAI gym tasks, achieving or outperforming state of the art in every environment tested.
  arXiv  Detail & Related papers  (2020-09-18T17:58:29Z)
• Deep Inverse Q-learning with Constraints [15.582910645906145]
  We introduce a novel class of algorithms that only needs to solve the MDP underlying the demonstrated behavior once to recover the expert policy. We show how to extend this class of algorithms to continuous state-spaces via function approximation and how to estimate a corresponding action-value function. We evaluate the resulting algorithms called Inverse Action-value Iteration, Inverse Q-learning and Deep Inverse Q-learning on the Objectworld benchmark.
  arXiv  Detail & Related papers  (2020-08-04T17:21:51Z)
• FACMAC: Factored Multi-Agent Centralised Policy Gradients [103.30380537282517]
  We propose FACtored Multi-Agent Centralised policy gradients (FACMAC) It is a new method for cooperative multi-agent reinforcement learning in both discrete and continuous action spaces. We evaluate FACMAC on variants of the multi-agent particle environments, a novel multi-agent MuJoCo benchmark, and a challenging set of StarCraft II micromanagement tasks.
  arXiv  Detail & Related papers  (2020-03-14T21:29:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.