AI
InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-Manipulation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 30/09/2026, 20:00:00 UTC
Briefing created: 02/10/2026, 11:21:58 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/10/2026, 06:18:24 UTC.
Briefings
InterEvolve addresses test-time evolution in humanoid loco-manipulation by repurposing existing controller skills without retraining. It achieves this via an object-aware forward-backward behavioral foundation model that converts frozen body states into new locomotion behaviors, combined with reward programs that guide LLM agents and numerical optimizers to compose motor competence. Experiments demonstrate that human-designed rewards often leave competence untapped, while InterEvolve-evolved programs successfully release and compose this skill for diverse tasks and physical units.
Key points
- InterEvolve uses an object-aware forward-backward (FB) behavioral foundation model to convert frozen body states into new locomotion behaviors at test time.
- Reward programs specify staged rewards with completion conditions and tunable constants, allowing an LLM agent and numerical optimizer to evolve strategies across parallel simulations.
Context and limitations
Coverage is limited to simulation environments and Unitree G1 physical units; long-horizon compositions and complex scenes are tested in simulation only.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 11:21:58 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiboilm7ujcpbhab7yxmw7puqcvj4sx2k54vcwelkl6perrqql7asy