Train dual-agent reinforcement learning with epistemic reward shaping in Jupyter notebooks.
npx skills add https://github.com/hrushi2501/rl-teacher --skill rl
Please help me install this Agent Skill. Skill: rl Source: https://github.com/hrushi2501/rl-teacher/tree/main/.agents/skills/rl Command: npx skills add https://github.com/hrushi2501/rl-teacher --skill rl
This Skill captures a reproducible, self-contained dual-agent reinforcement learning workflow that demonstrates epistemic reward shaping, enabling researchers to study how a teacher network can shape a student's learning via task-specific reward maps and per-task delta signals, improving generalization to unseen environments.
Open the epistemic_rl.ipynb notebook and run the training loop for the teacher-guided or baseline random conditions to reproduce PPOStudent and TeacherAgent interactions and export ONNX models.
High-intent search queries and answers about installing and using this skill.
Epistemic reward shaping uses a teacher network to output task-specific spatial reward maps, augmenting a student agent's rewards during training to improve zero-shot transfer to unseen environments.
Open the provided Jupyter notebook and run the training loop to execute dual PPO student networks and a teacher agent entirely from scratch, without requiring any external reinforcement learning libraries.
Yes, the notebook training loop automatically exports trained student and teacher models to ONNX format, enabling direct frontend visualization alongside JSON and episode logs for analysis.
Yes, the system trains across 10 gridworld tasks and evaluates generalization through zero-shot testing on 3 unseen gridworld layouts to quantify the transfer of learned policies.
The implementation mandates separate student network optimizers, per-task delta-based teacher rewards, no external RL libraries, deterministic notebook execution, and detailed logging to ensure reproducible experiments.
NVIDIA-TAOAutomate Grounding DINO training, evaluation, and ONNX export with PyTorch.
Simon-Copilot-StudioSegment objects in images using zero-shot point, box, or mask prompts.
thinking-machines-labAutomate knowledge distillation from teacher to student models via the Tinker API.
jackspaceOptimize high-throughput PPO+LSTM reinforcement learning training with vectorized environments.
