Downloads · 30 days
0
walker2d-selfanchored-v2.3
rijal028/walker2d-selfanchored-v2.3
walker2d-selfanchored-v2.3 is a reinforcement learning model from rijal028. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product.
Arsitektur model bipedal locomotion adaptif yang mengintegrasikan Instant Live Self-Anchoring: mengunci 2 frame output alami robot sesaat sebelum anomali sebagai regularizer topologi manifold, mencegah catastrophic fo…
Access
Public
Updated Aug 23, 2026
Repo size
968 KB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt968 KB · 100%
At a glance
- Task
- Reinforcement Learning
- Access
- Public
- Created
- Aug 23, 2026
- Updated
- Aug 23, 2026
- SHA
- be448ca1
Model card
From the Hugging Face model README
Walker2d Self-Anchored Autonomous Agent (v2.2)
Arsitektur model bipedal locomotion adaptif yang mengintegrasikan Instant Live Self-Anchoring: mengunci 2 frame output alami robot sesaat sebelum anomali sebagai regularizer topologi manifold, mencegah catastrophic forgetting secara murni pada level sinapsis neural network.
Ekosistem Modul Terintegrasi
- Actor Network (Trained Weights): Mengonsolidasikan respons dorongan angin badai (+140N) dan kelumpuhan sendi secara permanen.
- World Dynamics Model: Detektor kejutan fisik terkalibrasi (Threshold: 8.0, Persistence: 3 frame).
- Critic Reflection Engine: Simulator pencarian aksi darurat saat terjadi anomali baru.
- Dynamic Self-Anchor Engine: Regularisasi live backprop dengan rasio $1:2$ tanpa memerlukan modul tambahan (LoRA/Adapter sintetis).
Hasil Validasi Retensi (Episode 2 / Zero External Help)
- Total Steps: 1.000 steps penuh (Full Run) vs Baseline 367 steps (+633 steps).
- Total Jarak: +24.617 m (vs Baseline +7.294 m).
- Kecepatan Lari: Stabil di kisaran 3.0 – 4.9 m/s pasca-dorongan angin reda.
- Trigger Rate: Meredam kepanikan hingga mendekati nol (murni feedforward).
- Task
- Reinforcement Learning
- Created
- Aug 23, 2026
- Updated
- Aug 23, 2026