Policy-Space Diffusion for Physics-Based Character Animation
Bibliographic record
Abstract
Adapting motion to new contexts in digital entertainment often demands fast agile prototyping. State-of-the-art techniques use reinforcement learning policies for simulating the underlined motion in a physics engine. Unfortunately, policies typically fail on unseen tasks and it is too time-consuming to fine-tune the policy for every new morphological, environmental, or motion change. We propose a novel point of view on using policy networks as a representation of motion for physics-based character animation. Our policies are compact, tailored to individual motion tasks, and preserve similarity with nearby tasks. This allows us to view the space of all motions as a manifold of policies where sampling substitutes training. We obtain memory-efficient encoding of motion that leverages the characteristics of control policies such as being generative, and robust to small environmental changes. With this perspective, we can sample novel motions by directly manipulating weights and biases through a Diffusion Model. Our newly generated policies can adapt to previously unseen characters, potentially saving time in rapid prototyping scenarios. Our contributions include the introduction of Common Neighbor Policy regularization to constrain policy similarity during motion imitation training making them suitable for generative modeling; a Diffusion Model adaptation for diverse morphology; and an open policy dataset. The results show that we can learn non-linear transformations in the policy space from labeled examples, and conditionally generate new ones. In a matter of seconds, we sample a batch of policies for different conditions that show comparable motion fidelity metrics as their respective trained ones.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.006 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".