Efficient On-policy Visual-RL via Stochastic Decoupled Policy Gradient

We present the stochastic decoupled policy gradient (SDPG), a lightweight visual reinforcement learning (RL) method that trains diverse visuomotor control policies end-to-end within a few hours on a single NVIDIA RTX 4080 GPU. SDPG estimates policy gradients via random perturbations of trajectory rollouts, requiring orders of magnitude fewer batch-rendered environments and substantially reducing compute and memory overhead. On visual MuJoCo benchmarks, SDPG consistently outperforms baseline methods in training time, memory usage, and rewards. Finally, to support future research, we introduce a suite of realistic visual robotics benchmarks spanning dexterous manipulation, challenging locomotion, and demonstrate effective sim-to-real transfer on physical hardware.

Sample-Based Hybrid Mode Control: Asymptotically Optimal Switching of Algorithmic and Non-Differentiable Control Modes

This paper investigates a sample-based solution to the hybrid mode control problem across non-differentiable and algorithmic hybrid modes. Our approach reasons about a set of hybrid control modes as an integer-based optimization problem where we select what mode to apply, when to switch to another mode, and the duration for which we are in a given control mode. A sample-based variation is derived to efficiently search the integer domain for optimal solutions. We find our formulation yields strong performance guarantees that can be applied to a number of robotics-related tasks. In addition, our approach is able to synthesize complex algorithms and policies to compound behaviors and achieve challenging tasks. Last, we demonstrate the effectiveness of our approach in real-world robotic examples that require reactive switching between long-term planning and high-frequency control.

Accelerating Visual-Policy Learning through Parallel Differentiable Simulation

In this work, we propose a computationally efficient algorithm for visual policy learning that leverages differentiable simulation and first-order analytical policy gradients. Our approach decouple the rendering process from the computation graph, enabling seamless integration with existing differentiable simulation ecosystems without the need for specialized differentiable rendering software. This decoupling not only reduces computational and memory overhead but also effectively attenuates the policy gradient norm, leading to more stable and smoother optimization. We evaluate our method on standard visual control benchmarks using modern GPU-accelerated simulation. Experiments show that our approach significantly reduces wall-clock training time and consistently outperforms all baseline methods in terms of final returns. Notably, on complex tasks such as humanoid locomotion, our method achieves a $4\times$ improvement in final return, and successfully learns a humanoid running policy within 4 hours on a single GPU.

Is Bellman Equation Enough for Learning Control?

The Bellman equation and its continuous-time counterpart, the Hamilton-Jacobi-Bellman (HJB) equation, serve as necessary conditions for optimality in reinforcement learning and optimal control. While the value function is known to be the unique solution to the Bellman equation in tabular settings, we demonstrate that this uniqueness fails to hold in continuous state spaces. Specifically, for linear dynamical systems, we prove the Bellman equation admits at least $\binom{2n}{n}$ solutions, where $n$ is the state dimension. Crucially, only one of these solutions yields both an optimal policy and a stable closed-loop system. We then demonstrate a common failure mode in value-based methods: convergence to unstable solutions due to the exponential imbalance between admissible and inadmissible solutions. Finally, we introduce a positive-definite neural architecture that guarantees convergence to the stable solution by construction to address this issue.