Differentiable reimplementations of the Atari VCS provide a complex, fully known ground-truth system for testing gradient-based explainable AI methods.
hub
Brax–a differentiable physics engine for large scale rigid body simulation
29 Pith papers cite this work. Polarity classification is still indexing.
29
Pith papers citing it
hub tools
citation-role summary
background 2
method 1
citation-polarity summary
representative citing papers
jaxipm is the first GPU-batched IPOPT solver in JAX using heterogeneous iteration fusion and iteration-level batching, delivering up to 32.85x higher throughput than standard IPOPT on quadrotor NMPC benchmarks.
HPO enables unbiased policy optimization in hybrid action spaces by mixing differentiable simulation gradients with score-function estimates, outperforming PPO as continuous dimensions increase.
A quality-aware exploration method using return-conditioned sigmoid scheduling and per-agent RSQ metrics achieves top-tier returns on seven cooperative MARL benchmarks.
RAPO uses a dual robust RL formulation with trajectory-level adversarial networks and model-level Boltzmann reweighting over dynamics ensembles to improve policy resilience and out-of-distribution generalization while keeping the problem tractable.
Closed-loop prompt-based translation with hierarchical verification and iterative repair produces equivalent high-performance RL environments across five cases including new TCGJax.
Dual-stream excitatory/inhibitory networks trained with modulo error routing achieve 96.7% MNIST and 61.7% CIFAR-10 accuracy plus competitive RL performance, revealing task-dependent credit-assignment bottlenecks under Dale's principle.
Analytic Policy Gradients enable exact gradient computation via backpropagation through simulation for differentiable continuous control, with segmented backprop to mitigate degradation on long-horizon tasks.
UniLab is a CPU/GPU heterogeneous system for robot RL training using MuJoCoUni and MotrixSim backends that reports 3-10x end-to-end efficiency improvements and cross-platform compatibility beyond CUDA.
A vision-language framework generates text-based rigid-body scene configurations from videos using motion reasoning and optical flow, reporting 0.30 IoU on CLEVRER (7x over baselines) and transfer to 235 real videos.
ARC-RL is a new suite of four MuJoCo continuous-control environments featuring game-inspired hexapod and quadruped morphologies, a single closed-form multi-component reward function, CPG demonstrators, and empirical comparisons of online and offline-to-online RL algorithms.
Recasts sampling-based nonconvex optimization as smoothed gradient descent to obtain non-asymptotic convergence guarantees and introduces the DIDA annealed algorithm that converges to the global optimum.
RigidFormer learns mesh-free rigid dynamics from point clouds using object-centric anchors, Anchor-Vertex Pooling, Anchor-based RoPE, and differentiable Kabsch alignment to enforce rigidity.
Neural Control uses adjoint differentiation of equilibrium conditions to compute trajectory-dependent proxy gradients for history-dependent implicit models in deformable object manipulation.
Tempered sequential Monte Carlo samples from a Boltzmann-tilted distribution over controllers to optimize trajectories and policies under differentiable dynamics.
Framework using parameterized Signal Temporal Logic specifications to shape rewards for PPO-based RL, yielding tighter velocity tracking and more stable training than hand-crafted rewards on Barkour quadruped in MuJoCo simulation.
MoRE improves robot policy success rates by 44 percentage points by distilling mode redirection into weights, matching filtered retraining performance without inference overhead.
COP-Q uses Cholesky-Ordered Projection on joint Q-values to incorporate inter-objective covariance, preserving safety conservatism while improving sample efficiency in robot locomotion and navigation tasks.
SWIM is a single-instance imitation method for learning and generalizing physically simulated swimming motions to new environments, bodies, and styles.
Embodied AI requires query-conditioned world models that select the simplest physical abstraction sufficient to answer intervention queries.
MuJoCoUni introduces BatchEnvPool, a C++/pybind11-based executor providing persistent batched stateful MuJoCo environments with domain randomization, sensor queries, and short stepping.
OrbiSim builds a differentiable physics engine from world models to support gradient-based policy optimization and contact modeling in robotics.
Recurrent RL policies can have their hidden states aligned with PMP co-states through a derived loss, yielding robust performance on partially observable control tasks.
QDHUAC is a distributional, target-free QD-RL method that enables stable high-UTD training and competitive performance on Brax locomotion tasks using far fewer environment steps than prior approaches.
citing papers explorer
-
A Differentiable Atari VCS:A Complex, Fully Known Ground Truth for Explainable AI
Differentiable reimplementations of the Atari VCS provide a complex, fully known ground-truth system for testing gradient-based explainable AI methods.
-
Scaling Nonlinear Optimization: Many Problems One GPU
jaxipm is the first GPU-batched IPOPT solver in JAX using heterogeneous iteration fusion and iteration-level batching, delivering up to 32.85x higher throughput than standard IPOPT on quadrotor NMPC benchmarks.
-
Policy Optimization in Hybrid Discrete-Continuous Action Spaces via Mixed Gradients
HPO enables unbiased policy optimization in hybrid action spaces by mixing differentiable simulation gradients with score-function estimates, outperforming PPO as continuous dimensions increase.
-
Quality-Aware Exploration Budget Allocation for Cooperative Multi-Agent Reinforcement Learning
A quality-aware exploration method using return-conditioned sigmoid scheduling and per-agent RSQ metrics achieves top-tier returns on seven cooperative MARL benchmarks.
-
Robust Adversarial Policy Optimization Under Dynamics Uncertainty
RAPO uses a dual robust RL formulation with trajectory-level adversarial networks and model-level Boltzmann reweighting over dynamics ensembles to improve policy resilience and out-of-distribution generalization while keeping the problem tractable.
-
Automatic Generation of High-Performance RL Environments
Closed-loop prompt-based translation with hierarchical verification and iterative repair produces equivalent high-performance RL environments across five cases including new TCGJax.
-
Diffusing Blame: Task-Dependent Credit Assignment in Biologically Plausible Dual-Stream Networks
Dual-stream excitatory/inhibitory networks trained with modulo error routing achieve 96.7% MNIST and 61.7% CIFAR-10 accuracy plus competitive RL performance, revealing task-dependent credit-assignment bottlenecks under Dale's principle.
-
Backpropagating Through Simulation: Analytic Policy Gradients for Sample and Learning Efficient Differentiable Continuous Control
Analytic Policy Gradients enable exact gradient computation via backpropagation through simulation for differentiable continuous control, with segmented backprop to mitigate degradation on long-horizon tasks.
-
UniLab: A Heterogeneous Architecture for Robot RL Beyond GPU-Dominant Paradigms
UniLab is a CPU/GPU heterogeneous system for robot RL training using MuJoCoUni and MotrixSim backends that reports 3-10x end-to-end efficiency improvements and cross-platform compatibility beyond CUDA.
-
$\Delta$ynamics: Language-Based Representation for Inferring Rigid-Body Dynamics From Videos
A vision-language framework generates text-based rigid-body scene configurations from videos using motion reasoning and optical flow, reporting 0.30 IoU on CLEVRER (7x over baselines) and transfer to 235 real videos.
-
ARC-RL: A Reinforcement Learning Playground Inspired by ARC Raiders
ARC-RL is a new suite of four MuJoCo continuous-control environments featuring game-inspired hexapod and quadruped morphologies, a single closed-form multi-component reward function, CPG demonstrators, and empirical comparisons of online and offline-to-online RL algorithms.
-
Global Convergence of Sampling-Based Nonconvex Optimization through Diffusion-Style Smoothing
Recasts sampling-based nonconvex optimization as smoothed gradient descent to obtain non-asymptotic convergence guarantees and introduces the DIDA annealed algorithm that converges to the global optimum.
-
RigidFormer: Learning Rigid Dynamics using Transformers
RigidFormer learns mesh-free rigid dynamics from point clouds using object-centric anchors, Anchor-Vertex Pooling, Anchor-based RoPE, and differentiable Kabsch alignment to enforce rigidity.
-
Neural Control: Adjoint Learning Through Equilibrium Constraints
Neural Control uses adjoint differentiation of equilibrium conditions to compute trajectory-dependent proxy gradients for history-dependent implicit models in deformable object manipulation.
-
Tempered Sequential Monte Carlo for Trajectory and Policy Optimization with Differentiable Dynamics
Tempered sequential Monte Carlo samples from a Boltzmann-tilted distribution over controllers to optimize trajectories and policies under differentiable dynamics.
-
Learning Gait-Aware Quadruped Locomotion with Temporal Logic Specifications
Framework using parameterized Signal Temporal Logic specifications to shape rewards for PPO-based RL, yielding tighter velocity tracking and more stable training than hand-crafted rewards on Barkour quadruped in MuJoCo simulation.
-
Behavior Uncloning: Distilling Mode Redirection into Policy Weights without Inference-Time Steering
MoRE improves robot policy success rates by 44 percentage points by distilling mode redirection into weights, matching filtered retraining performance without inference overhead.
-
COP-Q: Safety-First Reinforcement Learning for Robot Control via Cholesky-Ordered Projection
COP-Q uses Cholesky-Ordered Projection on joint Q-values to incorporate inter-objective covariance, preserving safety conservatism while improving sample efficiency in robot locomotion and navigation tasks.
-
SWIM: Single-Instance Whole-Body Imitation for swiMming
SWIM is a single-instance imitation method for learning and generalizing physically simulated swimming motions to new environments, bodies, and styles.
-
Physically Viable World Models: A Case for Query-Conditioned Embodied AI
Embodied AI requires query-conditioned world models that select the simplest physical abstraction sufficient to answer intervention queries.
-
MuJoCoUni:Persistent Batched Runtime Primitives for MuJoCo
MuJoCoUni introduces BatchEnvPool, a C++/pybind11-based executor providing persistent batched stateful MuJoCo environments with domain randomization, sensor queries, and short stepping.
-
OrbiSim: World Models as Differentiable Physics Engines for Embodied Intelligence
OrbiSim builds a differentiable physics engine from world models to support gradient-based policy optimization and contact modeling in robotics.
-
Neural Co-state Policies: Structuring Hidden States in Recurrent Reinforcement Learning
Recurrent RL policies can have their hidden states aligned with PMP co-states through a derived loss, yielding robust performance on partially observable control tasks.
-
Distributional Value Estimation Without Target Networks for Robust Quality-Diversity
QDHUAC is a distributional, target-free QD-RL method that enables stable high-UTD training and competitive performance on Brax locomotion tasks using far fewer environment steps than prior approaches.
-
Real-to-Sim for Highly Cluttered Environments via Physics-Consistent Inter-Object Reasoning
A differentiable optimization pipeline uses a contact graph and rigid-body simulation to jointly refine object poses and physical properties, producing physically valid 3D scene reconstructions from single-view RGB-D observations for cluttered environments.
-
Does "Do Differentiable Simulators Give Better Policy Gradients?'' Give Better Policy Gradients?
In policy gradient RL, careful variance control and simple estimator switching frequently outperform explicit discontinuity detection even when using differentiable simulators.
-
GRaD-Nav++: Vision-Language Model Enabled Visual Drone Navigation with Gaussian Radiance Fields and Differentiable Dynamics
GRaD-Nav++ combines 3D Gaussian Splatting simulation and differentiable RL to train an onboard VLA policy that achieves 50-83% success on language-guided drone navigation tasks in simulation and real hardware.
-
Integrating Mechanistic and Data-Driven Models for Neurological Disorders through Differentiable Programming
This perspective paper categorizes hybrid architectures for combining mechanistic and data-driven models using residual learning, Neural ODEs, and solver-in-the-loop to model neurological disorder progression.
-
Chrono-Gymnasium: An Open-Source, Gymnasium-Compatible Distributed Simulation Framework
Chrono-Gymnasium is a Ray-based distributed wrapper that adds a Gymnasium interface to Project Chrono simulations, demonstrated on RL navigation and Bayesian lander optimization.