How do we make RL-powered humanoid robots safe without sacrificing performance? In our latest #CoRL2026# paper, Constrained Whole-Body Tracking for Humanoid Robots, we introduce safety filters that enforce kinematic and dynamic constraints—even those never encountered during training. The approach minimally interferes with the underlying RL policy and requires *no retraining*. At CoRL in Austin? Come meet @danielpmorton at Poster Session 2, Tuesday evening, #204#! Project: Paper: Work led by @danielpmorton, in collaboration with Pranit Mohnot #HumanoidRobotics# #ReinforcementLearning# #RobotLearning# #PhysicalAISafety# #Robotics# @StanfordAILab @StanfordASL

TL;DR: A new pipeline automatically builds 5,545 RL training tasks for coding agents using only source code itself, no issues or commit history needed, and it prioritizes quality over quantity. Title: CodeMidas: Scaling Agentic Coding RL Environments from Code Itself URL: Key points 🏗️ Auto-builds 5,545 tasks from 3,185 repos across 23 languages and 15 domains 🧪 Generates verifiers via execution-grounded tests, running the reference solution to record expected outputs 🛡️ Three-stage filtering: leakage checks, agent-solution agreement, and rollout difficulty filtering 📈 Big gains after RL training: DeepSWE +11.7pt, ProgramBench +17.0pt, Terminal-Bench +8.5pt 🔍 5k filtered tasks consistently beat 8k unfiltered tasks, proving quality beats quantity 🧠 Trained agents explore more and self-verify more, and these behaviors transfer to external benchmarks I like how simple and practical the core idea is: you don't need issue trackers or dev history to build RL environments, just the code. #CodingAgent# #ReinforcementLearning#

AGIBOT officially open-sourced AGIBOT WORLD 2026 Theme 3: Reinforcement Learning, a real-world dataset designed to advance reinforcement learning for embodied AI. Built for real-world robot learning, the dataset captures more than expert demonstrations — it captures successes, failures, and human corrections across the path from demonstration to deployment. • 14 real-world tasks • 11,430 trajectories • Fine-grained annotations for progress, mistakes & interference • Human-in-the-loop correction trajectories Learn from success. Learn from failure. Learn how to improve. 📖 Explore the project: 🤗 Access the dataset: #AGIBOT# #EmbodiedAI# #ReinforcementLearning# #PhysicalAI# #Robotics#

Today we're releasing @nvidia AlpaGym, our new open-source reinforcement learning (RL) framework for end-to-end autonomous driving. A key challenge for #Physical# #AI# is enabling policies to learn from the consequences of their actions. While supervised learning can teach a model to imitate behavior, robust autonomy ultimately requires learning through interaction with the environment. AlpaGym enables exactly that. Built on top of: - AlpaSim: our high-fidelity closed-loop autonomous driving simulator - Cosmos-RL: NVIDIA's distributed RL training and rollout infrastructure AlpaGym provides the glue that connects simulation, training, and driving policies into a scalable framework for post-training autonomous vehicle models in closed loop. With AlpaGym, researchers and developers can: ✅ Train end-to-end driving policies using reinforcement learning ✅ Run large-scale closed-loop simulations ✅ Experiment with new reward functions, policy architectures, and training strategies ✅ Benchmark models on public leaderboards 📖 Learn how it works: 💻 GitHub: 🏆 Open Challenges: - AlpaSim Closed-Loop E2E Driving Challenge: - Physical AI AV Reasoning Challenge: Learn more about the #Alpamayo# open platform: #PhysicalAI# #AutonomousDriving# #ReinforcementLearning# #Robotics# #OpenSource# #NVIDIA# #MachineLearning# @NVIDIADRIVE @NVIDIAAI

[𝐏𝐑𝐎𝐃𝐔𝐂𝐓 𝐔𝐏𝐃𝐀𝐓𝐄] 🇮🇹 🤖 #GENE01# designed from scratch and sent to batch production. Two scalable lower bodies. #PhysicalAI# deployed for motor control and world-action modeling. All in three months. @G_Bionics is running. 🎥 Enjoy the video below: blind #walking# and #running#, enabled by #reinforcementlearning# deployed across different robot platforms. 🙏 Kudos to the entire company for making these extraordinary results a reality. Stay tuned. The best is yet to come. #GenerativeBionics# #HumanoidRobotics# #AI# #Europe# #Innovation# #Startup# #DeepTech# #AMD# #PhysicalAI#

What should a search agent learn from when deciding whether to search? 🔎 Search When Needed, Proceed When Reliable: Metacognitive Reliability Supervision for Efficient Agentic Search ❓ What goes wrong when the reward is just final-answer correctness? 💡 A lucky guess, a correct answer with needless searches, and a wrong answer after useful retrieval all look alike, so the reward says little about each search decision. That leads to under-searching on unreliable states and over-searching on reliable ones. ❓ What does this paper reward instead? 💡 The estimated reliability of the intermediate conclusions the model writes at each step. Uncertainty is the mean negative log-likelihood of the tokens, and a mapping fitted beforehand on sampled data turns it into a reliability score. ❓ How are search and no-search decisions scored? 💡 ・If the agent proceeds without searching, the score is its current reliability minus a calibration reference ・If it searches, the score is reliability with the retrieved evidence minus reliability without it The average becomes a process reward added to GRPO. Probing is used only in training, so inference has no extra cost. ❓ How well does it work? 💡 With Qwen2.5-7B it averages 69.3 CEM over seven QA benchmarks, versus 67.2 for the strongest baseline, HiPRAG, with 1.64 searches per question (6.8% fewer). Multi-hop searches rise about 6.6%, and training is about 14.6% slower per step. #AgenticRAG# #ReinforcementLearning#

Do pretraining, SFT, and RL really need to be separate stages? Title: Training Foundation Models with a Single Algorithm URL: This paper shows a framework that runs everything from pretraining to RL together, with one likelihood-based loss and one trainer. Highlights 🧩 RL is likelihood fitting too The MaxRL gradient has the same form as MLE on the distribution of successful completions. So pretraining, midtraining, SFT, and RL differ only in data mixture and timing, and staged training is a special case. 📈 It beats staged training On reasoning puzzles with a 124M model, the best unified configuration beat staged training on all three evaluations. Staged training degrades when midtraining runs past its best budget. 🛡️ Forgets less, learns the next task better With a 0.76B model, held-out corpus loss went below the base model, while staged training raised it. After math, coding RL reached a higher final accuracy for every unified model than for the staged one. Experiments are small-scale, but I find the view of training as "which data, when, in what proportion" compelling. #FoundationModels# #ReinforcementLearning#

🎮 A game-playing AI that can walk around and drive a car in a commercial game it has never trained on - and it learned to play without ever touching a real game, only synthetic environments. Title: Player-1: Learning to Play from Synthetic Experience URL: Here are 3 highlights. 🕹️ A game-specific action space Relative pointer motion handles camera control, "sustained inputs" let a key stay held down, and "composite actions" (like holding W while tapping jump) let the model express the kind of coordinated controls games actually need. 🏗️ RL training across 16 synthetic environments 16 verifiable 2D/3D environments cover aiming, combat, driving, navigation, and UI interaction, with appearance, camera, and control sensitivity all randomized together to avoid overfitting to any single game. 🚀 Generalizing to games it never trained on On Minecraft (MineDojo-300, outside the RL training set), success jumps from 9.21% to 38.33%; on ViZDoom, scaled return improves 83.7%. In the unseen commercial game Watch Dogs 2, it walks, approaches a vehicle, gets in, and drives through the city. A nice finding on top: scaling synthetic environments from 2 to 16 steadily improves generalization to unseen games, which is a practical takeaway for building future game agents. #GameAI# #ReinforcementLearning#

🧩 Agent training environments are often surprisingly idealized: tools always respond cleanly, users always give clear instructions. But how well does an agent raised on that assumption hold up against real timeouts, errors, and sudden mid-task corrections from a real user? That's the question this paper takes on. The authors mined over 24,000 real tool failures across 557 MCP servers, distilling them into 8 failure families and 1,170 "recovery skills." They then teach an agent to reflect from three angles - recovering right after a failure, replanning from scratch, and responding to natural-language corrections from a user - and sharpen that ability further with reinforcement learning on a mix of noisy and clean rollouts. That's Agent-Reflex. The results back it up: average accuracy across 5 long-horizon tool-use benchmarks jumps from 31.6% for the base model to 58.6%, and even under injected noise the model holds at 45.9%. Baking the ability to recover from failure into the model's parameters, rather than relying on prompting, is what makes the difference. Title: Agent-Reflex: Teaching Language Agents to Act Reflectively URL: #LLMAgents# #ReinforcementLearning#

🎮 A game-playing AI that can walk around and drive a car in a commercial game it has never trained on - and it learned to play without ever touching a real game, only synthetic environments. Title: Player-1: Learning to Play from Synthetic Experience URL: Here are 3 highlights. 🕹️ A game-specific action space Relative pointer motion handles camera control, "sustained inputs" let a key stay held down, and "composite actions" (like holding W while tapping jump) let the model express the kind of coordinated controls games actually need. 🏗️ RL training across 16 synthetic environments 16 verifiable 2D/3D environments cover aiming, combat, driving, navigation, and UI interaction, with appearance, camera, and control sensitivity all randomized together to avoid overfitting to any single game. 🚀 Generalizing to games it never trained on On Minecraft (MineDojo-300, outside the RL training set), success jumps from 9.21% to 38.33%; on ViZDoom, scaled return improves 83.7%. In the unseen commercial game Watch Dogs 2, it walks, approaches a vehicle, gets in, and drives through the city. A nice finding on top: scaling synthetic environments from 2 to 16 steadily improves generalization to unseen games, which is a practical takeaway for building future game agents. #GameAI# #ReinforcementLearning#