A2C PandaReachDense-v3
Overview
An Advantage Actor Critic (A2C) agent trained to control a Franka Panda robotic arm using Panda-Gym.
The objective is to move the robot end-effector towards a desired target position.
Training Details
- Algorithm: A2C
- Framework: Stable-Baselines3
- Environment: PandaReachDense-v3
- Policy: MultiInputPolicy
- Timesteps: 1,000,000
- Parallel environments: 4
- Observation normalization: Enabled
- Reward normalization: Enabled
- Random seed: 42
Evaluation Results
- Evaluation episodes: 100
- Mean reward: -0.2246
- Standard deviation: 0.1156
- Certification score (mean - std): -0.3402
- Required certification score: -3.5
- Status: Passed
Files
- a2c-PandaReachDense-v3.zip: Trained model
- vec_normalize.pkl: Normalization statistics
- train.py: Training implementation
- requirements.txt: Dependencies
- results.json: Evaluation results
- replay.mp4: Robot demonstration
- Downloads last month
- 24