yardenas/safe-sim2real
Nov 2024 – Oct 2025
12 Months active
Languages Used
JAXPythonYAMLyamlXMLPDFTOMLShell
Technical Skills
Algorithm DesignAlgorithm ImplementationAlgorithm OptimizationBraxConfiguration ManagementDeep Learning
Yarden As
PROFILE
Yarden As
Yarden As worked extensively on the safe-sim2real repository, building robust reinforcement learning workflows that bridge simulation and real-world robotics. Over twelve months, Yarden engineered modular pipelines for data collection, online learning, and policy optimization, integrating algorithms like PPO, SAC, and MBPO with safety-aware control and vision-based planning. Using Python, JAX, and MuJoCo, Yarden refactored core components for maintainability, implemented hardware integrations for RCCar and Franka, and enhanced experiment reproducibility through improved logging, checkpointing, and visualization. The work demonstrated technical depth in algorithm development, system integration, and code quality, resulting in a scalable, reliable platform for safe robotics experimentation and deployment.
Overall Statistics
Feature vs Bugs
74%Features
Repository Contributions
499Total
Bugs
74
Commits
499
Features
211
Lines of code
69,734
Activity Months12
Your Network
22 people
Work History
October 2025
21 Commits • 11 Features
Oct 1, 2025October 2025 delivered substantial improvements to the safe-sim2real workflow, with a focus on reliability, online operation, and learning stability. Key features include Franka experiments improvements (increasing UTD), RCCar enhancements with RAe integration and 100% online mode, humanoid development including standup and nonepisodic tasks with termination handling, and notable algorithmic/training enhancements (pessimistic Q parameter and SAC alignment) plus loading behavior improvements during actor-critic training. Documentation updates and code cleanup were completed to improve maintainability and developer experience.
21 Commits • 11 Features
Oct 1, 2025October 2025 delivered substantial improvements to the safe-sim2real workflow, with a focus on reliability, online operation, and learning stability. Key features include Franka experiments improvements (increasing UTD), RCCar enhancements with RAe integration and 100% online mode, humanoid development including standup and nonepisodic tasks with termination handling, and notable algorithmic/training enhancements (pessimistic Q parameter and SAC alignment) plus loading behavior improvements during actor-critic training. Documentation updates and code cleanup were completed to improve maintainability and developer experience.
October 2025
September 2025
11 Commits • 3 Features
Sep 1, 2025Month: 2025-09 – yardenas/safe-sim2real\n\nKey features delivered:\n- RL improvements and safety enhancements: LBSGD penalizer, pure exploration in MBPO, non-episodic training mode, Actsafe safety configuration, and on-policy training step refactor for modularity.\n- Simulation Visualization Improvements: render the box as a red semi-transparent sphere and enable use_ball in Franka simulation.\n- Documentation and Repository Maintenance: updated installation/docs commands, added SPIDR citation, and cleaned obsolete configs to improve onboarding.\n\nMajor bugs fixed:\n- No critical defects reported. Stability and safety enhancements reduce edge-case failures: hardened Actsafe safety configuration and modular on-policy training refactor.\n\nOverall impact and accomplishments:\n- Delivered safer, data-efficient RL workflows; improved simulation fidelity and debugging; streamlined onboarding; reduced maintenance burden through code modularity and config cleanup.\n\nTechnologies/skills demonstrated:\n- Reinforcement Learning (MBPO, LBSGD, on-policy training)\n- Safety engineering (Actsafe)\n- Simulation visualization and debugging\n- Python, Git/versioned collaboration, documentation best practices
September 2025
11 Commits • 3 Features
Sep 1, 2025Month: 2025-09 – yardenas/safe-sim2real\n\nKey features delivered:\n- RL improvements and safety enhancements: LBSGD penalizer, pure exploration in MBPO, non-episodic training mode, Actsafe safety configuration, and on-policy training step refactor for modularity.\n- Simulation Visualization Improvements: render the box as a red semi-transparent sphere and enable use_ball in Franka simulation.\n- Documentation and Repository Maintenance: updated installation/docs commands, added SPIDR citation, and cleaned obsolete configs to improve onboarding.\n\nMajor bugs fixed:\n- No critical defects reported. Stability and safety enhancements reduce edge-case failures: hardened Actsafe safety configuration and modular on-policy training refactor.\n\nOverall impact and accomplishments:\n- Delivered safer, data-efficient RL workflows; improved simulation fidelity and debugging; streamlined onboarding; reduced maintenance burden through code modularity and config cleanup.\n\nTechnologies/skills demonstrated:\n- Reinforcement Learning (MBPO, LBSGD, on-policy training)\n- Safety engineering (Actsafe)\n- Simulation visualization and debugging\n- Python, Git/versioned collaboration, documentation best practices
August 2025
25 Commits • 15 Features
Aug 1, 2025August 2025 monthly summary for yardenas/safe-sim2real: Delivered a cohesive set of data, vision, and robotics features that accelerate experimentation and broaden hardware applicability, while strengthening safety and maintenance. Key data and perception features include RGB image preprocessing, loading auxiliary data, and offline data collection to enable faster iteration and richer datasets. Vision-driven planning and learning capabilities were expanded through SPiDR from Vision, MBPO with a vision wrapper, and MBPO + Vision, complemented by SOOPER integration between the SOOPER framework and Vision. Hardware and control enhancements broaden the platform’s reach with RCCar baseline and Franka-related updates. Documentation and logging improvements improve traceability and onboarding. Major bug fixes address normalization, safety policy handling, and core/vision integration stability, contributing to a more reliable development and experimentation environment.
25 Commits • 15 Features
Aug 1, 2025August 2025 monthly summary for yardenas/safe-sim2real: Delivered a cohesive set of data, vision, and robotics features that accelerate experimentation and broaden hardware applicability, while strengthening safety and maintenance. Key data and perception features include RGB image preprocessing, loading auxiliary data, and offline data collection to enable faster iteration and richer datasets. Vision-driven planning and learning capabilities were expanded through SPiDR from Vision, MBPO with a vision wrapper, and MBPO + Vision, complemented by SOOPER integration between the SOOPER framework and Vision. Hardware and control enhancements broaden the platform’s reach with RCCar baseline and Franka-related updates. Documentation and logging improvements improve traceability and onboarding. Major bug fixes address normalization, safety policy handling, and core/vision integration stability, contributing to a more reliable development and experimentation environment.
August 2025
July 2025
46 Commits • 22 Features
Jul 1, 2025July 2025: Built and validated end-to-end reinforcement learning workflows for safe-sim2real, including benchmarks, hardware integrations, robust training pipelines, and improved documentation. Implemented image observations in SAC, vision enhancements, and MBPO-related updates, while refactoring project structure to improve maintainability. These efforts enabled faster experiment cycles, more reliable models, and deeper real-world evaluation with RCCar and Franka hardware.
July 2025
46 Commits • 22 Features
Jul 1, 2025July 2025: Built and validated end-to-end reinforcement learning workflows for safe-sim2real, including benchmarks, hardware integrations, robust training pipelines, and improved documentation. Implemented image observations in SAC, vision enhancements, and MBPO-related updates, while refactoring project structure to improve maintainability. These efforts enabled faster experiment cycles, more reliable models, and deeper real-world evaluation with RCCar and Franka hardware.
June 2025
46 Commits • 19 Features
Jun 1, 2025June 2025 monthly summary for yardenas/safe-sim2real: Delivered substantial MBPO-based RL enhancements, reinforcing safety and data efficiency, with core MBPO implementation, PPO compatibility, SAC optimism, and online data integration. Implemented safety/terminations in the planning MDP and restored the best working version to stabilize training, alongside an optimizer state persistence feature to resume experiments. Expanded experimentation capabilities with RAE improvements, hard autoreset wrapper, and new go2 and Aloha features, plus hyperparameter scheduling experiments for faster convergence. Strengthened operations and maintainability through documentation and dependencies updates, training-time optimizations, on-demand import improvements, and robust error handling. Enhanced tooling integration (moviepy, WandB entity config, and parameterizable sooper filtering) to improve research workflows and reproducibility.
46 Commits • 19 Features
Jun 1, 2025June 2025 monthly summary for yardenas/safe-sim2real: Delivered substantial MBPO-based RL enhancements, reinforcing safety and data efficiency, with core MBPO implementation, PPO compatibility, SAC optimism, and online data integration. Implemented safety/terminations in the planning MDP and restored the best working version to stabilize training, alongside an optimizer state persistence feature to resume experiments. Expanded experimentation capabilities with RAE improvements, hard autoreset wrapper, and new go2 and Aloha features, plus hyperparameter scheduling experiments for faster convergence. Strengthened operations and maintainability through documentation and dependencies updates, training-time optimizations, on-demand import improvements, and robust error handling. Enhanced tooling integration (moviepy, WandB entity config, and parameterizable sooper filtering) to improve research workflows and reproducibility.
June 2025
May 2025
40 Commits • 20 Features
May 1, 2025May 2025 (yardenas/safe-sim2real) delivered a sequence of robust RL ecosystem enhancements, data pipeline improvements, and repo hygiene improvements that collectively accelerate experimentation, improve data quality, and enhance real-world transfer fidelity. Key features delivered include adding dictionary observations support in Ramu, integrating PPO for policy optimization, and a major refactor of the data collection and training pipelines to support episodic data collection. Plotting updates and reward refinements improved observability and guidance for tuning. Training pipeline updates, online learning capabilities, and checkpointing support bolster experimentation speed and resilience. Documentation updates and code cleanup further improve maintainability. Major bug fixes included reverting train-logging changes to restore expected behavior and removing DS_STORE artifacts to reduce noise. Overall, these changes increase model fidelity, reduce debugging time, and empower faster, more reliable iterations toward real-world deployment.
May 2025
40 Commits • 20 Features
May 1, 2025May 2025 (yardenas/safe-sim2real) delivered a sequence of robust RL ecosystem enhancements, data pipeline improvements, and repo hygiene improvements that collectively accelerate experimentation, improve data quality, and enhance real-world transfer fidelity. Key features delivered include adding dictionary observations support in Ramu, integrating PPO for policy optimization, and a major refactor of the data collection and training pipelines to support episodic data collection. Plotting updates and reward refinements improved observability and guidance for tuning. Training pipeline updates, online learning capabilities, and checkpointing support bolster experimentation speed and resilience. Documentation updates and code cleanup further improve maintainability. Major bug fixes included reverting train-logging changes to restore expected behavior and removing DS_STORE artifacts to reduce noise. Overall, these changes increase model fidelity, reduce debugging time, and empower faster, more reliable iterations toward real-world deployment.
April 2025
65 Commits • 30 Features
Apr 1, 2025April 2025 (2025-04) delivered a robust end-to-end evaluation workflow via the Evaluation System integration, expanded RCCar parameterization and global parameter updates for improved performance and configurability, and integrated Safety Gym environments with reliability hardening to enable stable experiments. The month also expanded the sensing suite with new sensors and introduced the Ramu component, and advanced PPO-based training with sim-to-real alignment to boost transfer of policies to hardware. Plotting utilities were added and plots updated for clearer diagnostics, while code quality improvements and bug fixes (environment reset handling, rendering/measurement reliability) improved overall stability. These efforts increased experimentation throughput, data quality, and alignment with real hardware, accelerating decision-making and deployment readiness.
65 Commits • 30 Features
Apr 1, 2025April 2025 (2025-04) delivered a robust end-to-end evaluation workflow via the Evaluation System integration, expanded RCCar parameterization and global parameter updates for improved performance and configurability, and integrated Safety Gym environments with reliability hardening to enable stable experiments. The month also expanded the sensing suite with new sensors and introduced the Ramu component, and advanced PPO-based training with sim-to-real alignment to boost transfer of policies to hardware. Plotting utilities were added and plots updated for clearer diagnostics, while code quality improvements and bug fixes (environment reset handling, rendering/measurement reliability) improved overall stability. These efforts increased experimentation throughput, data quality, and alignment with real hardware, accelerating decision-making and deployment readiness.
April 2025
March 2025
125 Commits • 49 Features
Mar 1, 2025March 2025 (2025-03) monthly summary for yardenas/safe-sim2real. Delivered end-to-end enhancements enabling faster experimentation, improved observability, and safer simulations. Key features expanded the platform's capabilities (DM Playground with privileged access, domain-randomized rendering, and extended sim-to-sim runs), along with governance and usability improvements (logging training metrics, updated params/UX, and streamlined workflows). Substantial RL and physics work advanced Go1 integration and DM scenarios, while stability and safety hardening reduced risk in production-like runs.
March 2025
125 Commits • 49 Features
Mar 1, 2025March 2025 (2025-03) monthly summary for yardenas/safe-sim2real. Delivered end-to-end enhancements enabling faster experimentation, improved observability, and safer simulations. Key features expanded the platform's capabilities (DM Playground with privileged access, domain-randomized rendering, and extended sim-to-sim runs), along with governance and usability improvements (logging training metrics, updated params/UX, and streamlined workflows). Substantial RL and physics work advanced Go1 integration and DM scenarios, while stability and safety hardening reduced risk in production-like runs.
February 2025
46 Commits • 16 Features
Feb 1, 2025February 2025: Delivered foundational code quality improvements, architectural decoupling, and feature-rich expansions across the safe-sim2real project. Achieved measurable business value by stabilizing the core pipeline, enabling reproducible experiments, and expanding agent capabilities and data visualization for faster decision-making.
46 Commits • 16 Features
Feb 1, 2025February 2025: Delivered foundational code quality improvements, architectural decoupling, and feature-rich expansions across the safe-sim2real project. Achieved measurable business value by stabilizing the core pipeline, enabling reproducible experiments, and expanding agent capabilities and data visualization for faster decision-making.
February 2025
January 2025
4 Commits • 2 Features
Jan 1, 20252025-01 monthly summary: Delivered core visualization capabilities for safe-sim2real, enabling rapid data interpretation and reproducibility. Key work includes adding Tueplots as a development dependency (pyproject.toml and poetry.lock updated) and implementing a plotting framework for simulation data and RCCar experiments, with scripts to load data from Weights & Biases, compute metrics, and generate visualizations using Seaborn/Matplotlib. RCCar plots received label improvements for clarity. No major bugs were fixed this month. These efforts improve decision support, accelerate experimentation cycles, and enhance the reproducibility and quality of published results.
January 2025
4 Commits • 2 Features
Jan 1, 20252025-01 monthly summary: Delivered core visualization capabilities for safe-sim2real, enabling rapid data interpretation and reproducibility. Key work includes adding Tueplots as a development dependency (pyproject.toml and poetry.lock updated) and implementing a plotting framework for simulation data and RCCar experiments, with scripts to load data from Weights & Biases, compute metrics, and generate visualizations using Seaborn/Matplotlib. RCCar plots received label improvements for clarity. No major bugs were fixed this month. These efforts improve decision support, accelerate experimentation cycles, and enhance the reproducibility and quality of published results.
December 2024
24 Commits • 8 Features
Dec 1, 2024December 2024 (2024-12) performance summary for yardenas/safe-sim2real: Delivered high-impact features and stability improvements for humanoid-enabled simulation workflows, with a clear focus on realism, safety, and maintainability. The work enabled end-to-end humanoid support, longer training horizons, and more realistic parameter alignment with real-world data, while also improving the code quality and governance around configuration tuning.
24 Commits • 8 Features
Dec 1, 2024December 2024 (2024-12) performance summary for yardenas/safe-sim2real: Delivered high-impact features and stability improvements for humanoid-enabled simulation workflows, with a clear focus on realism, safety, and maintainability. The work enabled end-to-end humanoid support, longer training horizons, and more realistic parameter alignment with real-world data, while also improving the code quality and governance around configuration tuning.
December 2024
November 2024
46 Commits • 16 Features
Nov 1, 2024November 2024 (yardenas/safe-sim2real): Delivered a focused set of features and robustness fixes to enable safer, faster experimentation and closer real-world alignment. Key work concentrated on real-environment evaluation, sim-to-real alignment, and broader deployment scenarios, complemented by stability and performance improvements across the training loop and hardware configuration. Resulted in a more reliable real-to-sim to real workflow, improved training stability, and clearer paths toward production-ready experiments.
November 2024
46 Commits • 16 Features
Nov 1, 2024November 2024 (yardenas/safe-sim2real): Delivered a focused set of features and robustness fixes to enable safer, faster experimentation and closer real-world alignment. Key work concentrated on real-environment evaluation, sim-to-real alignment, and broader deployment scenarios, complemented by stability and performance improvements across the training loop and hardware configuration. Resulted in a more reliable real-to-sim to real workflow, improved training stability, and clearer paths toward production-ready experiments.
Activity
Loading activity data...
Quality Metrics
Correctness83.6%
Maintainability85.2%
Architecture81.6%
Performance71.0%
AI Usage20.6%
Skills & Technologies
Programming Languages
CSVJAXMarkdownNumPyONNXPDFPythonShellTOMLTensorFlow
Technical Skills
API IntegrationAlgorithm ConfigurationAlgorithm DebuggingAlgorithm DesignAlgorithm DevelopmentAlgorithm ImplementationAlgorithm IntegrationAlgorithm OptimizationAlgorithm RefactoringBenchmark ConfigurationBenchmark DevelopmentBenchmark SetupBraxBuffer ManagementBug Fix
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline