Reproducible Debugging: Git, Downloads, ABI, and Evidence States
Locate installation, network, extension, and runtime failures by layer and sanitize evidence.
RESEARCH & ENGINEERING
Research & Engineering
Research discussions, engineering practice, and experimental reproduction, organized by category.
GPU Software Stacks and Correct PyTorch/JAX Timing
Separate drivers, toolkits, and framework runtimes; synchronize asynchronous benchmarks.
Python Engineering: pyproject, uv Locking, and Build Isolation
Separate runtime and build dependencies; diagnose installation through wheel compatibility and backend errors.
Paired Direct/Agent Evaluation and Progress Verification
Fix episode identity and budgets, separating task success, tool completion, and ablations.
OpenPI × LIBERO: Serving, Observations, and Action Contracts
Use pinned upstream code to check fields, image orientation, action queues, and a single request.
VLA Execution: Action Chunks, Latency, and Feedback
Separate action representation, generation, consumption, and scheduling; measure observation age.
Navigation Baselines and Migration: Auditing Papers, Code, and Assets
Audit versioned interfaces for NavDP, FLUX, NavIsaacLab, and ProtoMotions.
From Video to Simulation: Geometry, Scale, Contact, and Navigation
Connect COLMAP and OpenUSD while distinguishing visual reconstruction from physical assets.
Interfaces for Crowd Behavior, Body Motion, and Physical Control
Separate intent, paths, body references, and execution, with explicit motion/policy dependencies.
ROS 2 and Simulation: Clocks, TF, QoS, and Stale Actions
Use Jazzy documentation to check message compatibility, timestamps, and reset boundaries.
Social Navigation Benchmarks: Contracts and Metrics
Define a twelve-module architecture and separate privileged observations, safety, and social metrics.
Humanoid Robots: Control Dimensions, Symmetry, and Curriculum Recovery
Separate model coordinates from policy actions and validate contacts, resets, and recovery.
Simulation Timing, NavMesh Updates, and Profiling
Separate physics, control, sensing, and rendering clocks and isolate startup costs.
Minimal Experiments for Contact and Mecanum Wheels
Inspect collision proxies, roller axes, support envelopes, and wheel-speed conventions.
Robot Assets: CAD, USD, URDF, and Inertial Validation
Validate units, composition layers, joint frames, and physical properties in stages.
RL Implementation: Objectives, Masks, and Gradient Paths
Turn PPO/GAE, TD, centralized training, and representations into inspectable interfaces.
Autograd, Gradient Accumulation, and Training Recovery
Understand graph lifetime, inference modes, gradient accumulation, and checkpoint contracts.
PyTorch: Axes, Broadcasting, and Storage Layout
Define shape contracts and distinguish axis changes, reshaping, views, and copies.
A Reproducible RL Evaluation and Reporting Protocol
Align environment versions, training budgets, checkpoint selection, held-out testing, and aggregation.
Multi-Agent RL Benchmarks and Cooperative Evaluation
Cooperative, competitive, mixed-motive, partially observed, and cross-partner MARL evaluation.
Single-Agent RL Benchmarks and Metrics
Consolidate five reports across control, visual generalization, offline datasets, and embodied tasks.
RL Statistics, Confidence Intervals, and Tuning Budgets
Seeds versus episodes, power, bootstrap, IQM, and tuning leakage.
TD7 Policy Checkpoints: Separating Training and Evaluation
Candidate and deployed policies, assessment, and independent evaluation.
Embodied RL, LLMs/VLAs, and Research Design
Turn sample efficiency, adaptation, rewards, and timing into controlled research questions.
Simulators, World Models, and Planning
Experimental environments, learned dynamics, and internal rollout serve different roles.
Meta-RL: Task Distributions and Rapid Adaptation
Optimization, contextual inference, and recurrent adaptation across tasks.
Markov States, Partial Observations, and Recurrent Memory
Separate physical state, observation, history, and belief without assuming RNN sufficiency.
IPPO, COMA, and Multi-Agent Gradient Variance
Independent learning, centralized critics, counterfactual credit, and a bounded variance example.
Goal-Conditioned RL, UVFA, and HER
Goal-conditioned values, hindsight relabeling, and hierarchical-control interfaces.
MDP Homomorphisms and State–Action Abstraction
Reward and transition preservation, policy lifting, bisimulation, and approximate representations.
Options, SMDPs, and Intra-Option Learning
Temporal abstraction, duration-dependent discounting, and experience shared across options.
Potential-Based Reward Shaping
Telescoping returns, policy invariance, terminal potentials, and reward-scale caveats.
Maximum-Entropy RL, SAC, and Dual Temperature Updates
Merge five notes and correct temperature-gradient signs and the off-policy/offline distinction.
Reparameterization and Pathwise Gradients
Gaussian reparameterization, tanh density corrections, and the limits of pathwise differentiation.
Offline RL and Value Extrapolation Error
Value estimates outside data support, maximization backups, and error propagation.
Diffusion Policy: Distribution Modeling and Control Timing
A supplement to P08 on denoising time, action chunks, and the execution loop.
Behavior Cloning, Action Bins, and Multimodal Actions
Supervised objectives, closed-loop distribution shift, and regression, classification, and residual actions.
Eligibility Traces and TD(λ)
Forward lambda returns, backward traces, online equivalence conditions, and episode boundaries.
Policy Gradients, REINFORCE, and Surrogate Objectives
Derive the trajectory gradient and connect baselines, PPO clipping, and KL diagnostics.
On-Policy Sampling, Off-Policy Learning, and Replay
Merge duplicate notes and distinguish behavior policies, target policies, and distribution shift.
Contractions, Fixed Points, and Stochastic Approximation
Separate deterministic Bellman contraction from convergence conditions for noisy recursions.
Probability Measures, Empirical Measures, and RL Occupancy
Probability kernels, trajectory laws, stationary distributions, and discounted occupancy measures.
MDPs, Value Functions, and Bellman Errors
Bellman equations, MSBE, MSPBE, sampled TD errors, and double sampling.
Cross-Paper Comparison and Research Roadmap
Connections and controlled experiments across action generation, policy adaptation, execution, human motion, and navigation worlds.
Hexo and GitHub Pages: Building and Publishing While Preserving Sources
Dependency installation, static generation, browser verification, and source-preserving Git publication.
No published records in this category.