Research map / Reinforcement learning / Policy gradient and actor-critic
SVL: Goal-Conditioned Reinforcement Learning as Survival Learning
Franki NGUIMATSIA TIOFACK, Fabian Schramm, Théotime Le Hellard, Justin Carpentier
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace TD-learned value (HIQL) with survival value learning (HSVL) | hierarchical GCRL · antmaze-giant-navigate | success rate (%): 65 -> 74 | Improved |
| replace TD-learned value (HIQL) with survival value learning (HSVL) | hierarchical GCRL · humanoidmaze-giant-navigate | success rate (%): 12 -> 81 | Improved |
| replace TD value with survival value on visual tasks | hierarchical GCRL · visual-antmaze-giant-navigate | success rate (%): 6 -> 61 | Improved |
| replace TD value with survival value on stitch tasks | hierarchical GCRL · antmaze-large-stitch | success rate (%): 67 -> 31 | Hurt |
| replace TD value with survival value on stitch tasks | hierarchical GCRL · antmaze-medium-stitch | success rate (%): 94 -> 83 | Hurt |
| use HSVL instead of GCIQL on manipulation puzzle | GCIQL · puzzle-3x3-play | success rate (%): 95 -> 54 | Hurt |
| use survival critic (SVL) instead of contrastive critic (CRL re-eval), flat actor | flat DDPG+BC actor · humanoidmaze-giant-navigate | success rate (%): 7 -> 45 | Improved |
| use survival critic (SVL) instead of contrastive critic (CRL re-eval), flat actor | flat DDPG+BC actor · humanoidmaze-large-navigate | success rate (%): 42 -> 66 | Improved |
| use survival critic (SVL) instead of CRL re-eval, flat actor | flat DDPG+BC actor · antmaze-giant-navigate | success rate (%): 50 -> 43 | Hurt |
| add hierarchical extraction on top of SVL (ablation) | SVL · humanoidmaze-giant-navigate | success rate (%): 45 -> 81 | Improved |
| add hierarchical extraction on top of SVL (ablation) | SVL · humanoidmaze-medium/large-navigate | success rate (%): 66 -> 66 (large) | No clear effect |
| use PCS hazard instead of finite-horizon or PCH (ablation) | HSVL · antmaze/humanoidmaze navigate | success rate (%): 71 / 71 / 74 on antmaze-giant | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HIQL — HIQL: Offline Goal-Conditioned RL with Latent States as Actions (NeurIPS 2023) · better on 8 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Conservative Offline Goal-Conditioned Implicit V-Learning (ICML 2025 · similarity 0.89)
- Option-aware Temporally Abstracted Value for Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Physics-informed Value Learner for Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Risk-Conditioned Reinforcement Learning: A Generalized Approach for Adapting to Varying Risk Measures (AAAI 2024 · similarity 0.88)
- Probabilistic Subgoal Representations for Hierarchical Reinforcement Learning (ICML 2024 · similarity 0.87)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
