SPiDR¶
Zero-shot safety under domain randomisation
OnlineRobust safeBase · PPOClaims · dynamic shift
Constrained learning under domain randomisation with a pessimistic cost penalty.
Features¶
| Feature | Value |
|---|---|
| Group | Robust Safe Algorithms |
| Setting | Online, with a cost constraint |
| Family | Robust safe |
| Base algorithm | PPO |
| Claimed robustness | Dynamic shift |
| Shifted-env rollout | Yes |
| Adversarial network | No |
| Learned model | Yes |
| Training budget | Isaac Lab recipe |
| Original paper | As et al. SPiDR: A Simple Approach for Zero-Shot Safety in Sim-to-Real Transfer. NeurIPS, 2026. |
Mechanism¶
SPiDR trains a constrained policy under domain randomisation and adds a pessimistic cost penalty proportional to the disagreement of next-state predictions across sampled dynamics. It runs with the original solver and a joint-limit cost budget.
Run the method¶
./isaaclab.sh -p isaac/train.py --task g1 --method spidr --seed 0 --headless
./isaaclab.sh \
-p isaac/evaluate.py --task g1 --checkpoint runs/isaac/g1_spidr/seed0/model_899.pt --condition all --headless
| File | Path |
|---|---|
| Implementation | isaac/methods/spidr.py |
Training rollouts come from the method's own perturbed environment. The configuration files are explained in Train an algorithm, and the evaluation of the frozen checkpoint in Evaluation Protocol.
References¶
- As et al. SPiDR: A Simple Approach for Zero-Shot Safety in Sim-to-Real Transfer. NeurIPS, 2026.