add belief-guided grasp selection example

Add examples/manipulation/08_belief_grasp_selection.py. The tabletop hides
the object's orientation among three discrete pose hypotheses, each best
served by a different grasp. The agent maintains a belief over the poses,
picks the grasp with the highest expected success across that belief,
runs a Bayes update on every miss, and retries until one grasp lands.

Cover the loop with a smoke test that locks success, belief_update_count
>= 1, at least one grasp_miss failure, and that the final belief argmax
matches the true pose. Register a GIF maker and the rendered GIF.

Update README, examples/README.md, examples/manipulation/README.md,
docs/status.md, docs/plan.md, and docs/learning_paths.md to reflect the
new example, GIF, and test counts (24 examples, 23 GIFs, 66 tests).

Author: rsasaki0109

README.md

@@ -11,10 +11,10 @@ robotics loops rather than standalone algorithms.
 
 ## Current Status
 
-- 23 runnable examples
+- 24 runnable examples
 - 20 learning-path roadmap examples
-- 22 README GIFs generated from runnable examples
-- 65 smoke and regression tests
+- 23 README GIFs generated from runnable examples
+- 66 smoke and regression tests
 - Core dependencies only: `numpy` and `matplotlib`
 
 See `docs/status.md` for the implementation snapshot and `docs/plan.md` for
@@ -100,6 +100,10 @@ These GIFs are generated from the runnable examples, not separate animations.
 | --- | --- |
 | ![A target starts under a shelf, the robot detects a blocked grasp, pushes it into open space, and then picks it.](docs/assets/gifs/push_then_grasp.gif) | ![A suction sorter estimates per-object success probabilities, recovers from a suction miss, prepares the seal, retries, and sorts into bins.](docs/assets/gifs/probabilistic_suction_sorting.gif) |
 
+| Belief-guided grasp selection |
+| --- |
+| ![A grasp agent keeps a belief over three pose hypotheses, picks the grasp with highest expected success, misses, runs a Bayes update, and tries a different grasp.](docs/assets/gifs/belief_grasp_selection.gif) |
+
 ### Navigation and recovery
 
 | Reactive obstacle avoidance | Dynamic obstacle avoidance |

docs/assets/gifs/belief_grasp_selection.gif

@@ -27,6 +27,7 @@
 5. `examples/manipulation/05_object_search_and_pick.py`
 6. `examples/manipulation/06_push_then_grasp.py`
 7. `examples/manipulation/07_probabilistic_suction_sorting.py`
+8. `examples/manipulation/08_belief_grasp_selection.py`
 
 ## Embodied AI
 

docs/learning_paths.md

@@ -8,11 +8,11 @@ first-run experience small, fast, and failure-aware.
 
 The repository currently has:
 
-- 23 runnable examples
+- 24 runnable examples
 - 20 numbered learning-path examples
-- 3 extra examples outside the original learning-path roadmap
-- 22 generated README GIFs
-- 65 smoke and regression tests
+- 4 extra examples outside the original learning-path roadmap
+- 23 generated README GIFs
+- 66 smoke and regression tests
 - GitHub Actions CI for Python 3.10, 3.11, and 3.12
 - core dependencies limited to `numpy` and `matplotlib`
 - optional Gymnasium-style adapters for `GridWorld2D`,
@@ -85,6 +85,10 @@ Recent completed work:
   The agent wakes up with a bimodal pose belief, takes information actions
   toward the symmetric landmark, then switches one-way to goal navigation
   after the belief collapses. A GIF and smoke test cover the loop.
+- `examples/manipulation/08_belief_grasp_selection.py` was added. The agent
+  keeps a belief over three pose hypotheses, picks the grasp with the
+  highest expected success across that belief, and runs a Bayes update on
+  every miss until a grasp lands. GIF and smoke test cover the loop.
 
 The next agent should not redo those items. If any of them seem missing, first
 check the current branch and latest pulled commit.
@@ -231,25 +235,26 @@ Acceptance criteria:
 
 ### Priority 4: Add The Next Example Tier
 
-Goal: grow from 23 examples toward 30 examples without losing readability.
+Goal: grow from 24 examples toward 30 examples without losing readability.
 
 Already done from the previous tier:
 
 - `examples/navigation/10_localization_uncertainty_recovery.py` — pose
   uncertainty -> information action -> resume goal navigation.
+- `examples/manipulation/08_belief_grasp_selection.py` — pose belief ->
+  grasp choice -> failure update -> retry.
 
 Recommended next examples:
 
 | Priority | Example | Area | Loop |
 | --- | --- | --- | --- |
-| 1 | `examples/manipulation/08_belief_grasp_selection.py` | manipulation | pose belief -> grasp choice -> failure update |
-| 2 | `examples/manipulation/09_active_viewpoint_for_grasp.py` | manipulation | choose view -> reduce occlusion -> grasp |
-| 3 | `examples/embodied_ai/21_object_permanence_toy.py` | embodied AI | object disappears -> memory persists -> search |
-| 4 | `examples/embodied_ai/22_where_did_i_see_it.py` | embodied AI | memory query -> revisit place -> act |
-| 5 | `examples/world_models/23_model_error_recovery.py` | world model | prediction failure -> update model -> recover |
-| 6 | `examples/navigation/24_information_gain_navigation.py` | navigation | goal progress vs information gain |
-| 7 | `examples/manipulation/25_clear_path_before_pick.py` | manipulation | precondition failure -> clear obstacle -> pick |
-| 8 | `examples/navigation/27_multi_agent_avoidance.py` | navigation | observe agents -> avoid -> replan |
+| 1 | `examples/manipulation/09_active_viewpoint_for_grasp.py` | manipulation | choose view -> reduce occlusion -> grasp |
+| 2 | `examples/embodied_ai/21_object_permanence_toy.py` | embodied AI | object disappears -> memory persists -> search |
+| 3 | `examples/embodied_ai/22_where_did_i_see_it.py` | embodied AI | memory query -> revisit place -> act |
+| 4 | `examples/world_models/23_model_error_recovery.py` | world model | prediction failure -> update model -> recover |
+| 5 | `examples/navigation/24_information_gain_navigation.py` | navigation | goal progress vs information gain |
+| 6 | `examples/manipulation/25_clear_path_before_pick.py` | manipulation | precondition failure -> clear obstacle -> pick |
+| 7 | `examples/navigation/27_multi_agent_avoidance.py` | navigation | observe agents -> avoid -> replan |
 
 Selection rule:
 

docs/plan.md

@@ -5,10 +5,10 @@ see what exists, what is verified, and what should come next.
 
 ## Snapshot
 
-- Runnable examples: 23
+- Runnable examples: 24
 - Learning-path roadmap examples: 20
-- README GIFs: 22
-- Smoke and regression tests: 65
+- README GIFs: 23
+- Smoke and regression tests: 66
 - Core dependencies: `numpy`, `matplotlib`
 - Contributor extra: `pip install -e ".[dev]"`
 - CI: Python 3.10, 3.11, 3.12
@@ -30,7 +30,7 @@ goal-command example. The numbered 20-example roadmap is tracked in
 | --- | ---: | --- |
 | Runtime | 2 | smallest observe-act-observe loop, trace replay |
 | Navigation | 9 | reactive avoidance, dynamic obstacles, replanning, exploration, belief, active SLAM, MPC, recovery, localization recovery |
-| Manipulation | 7 | retry, reactive grasping, IK servo, moving target reaching, search, push recovery, suction sorting |
+| Manipulation | 8 | retry, reactive grasping, IK servo, moving target reaching, search, push recovery, suction sorting, belief grasp selection |
 | Embodied AI | 4 | controlled goals, memory, POMDP search, tiny VLA loop |
 | World models | 1 | action-conditioned dynamics, prediction error, model update, replanning |
 

docs/status.md

@@ -37,6 +37,7 @@ Run any example headless with its `--no-render` flag when available.
 | `manipulation/05_object_search_and_pick.py` | `python examples/manipulation/05_object_search_and_pick.py` | search viewpoint -> memory -> pick -> retry |
 | `manipulation/06_push_then_grasp.py` | `python examples/manipulation/06_push_then_grasp.py` | blocked grasp -> push world -> grasp |
 | `manipulation/07_probabilistic_suction_sorting.py` | `python examples/manipulation/07_probabilistic_suction_sorting.py` | suction miss -> update probability -> prepare -> sort |
+| `manipulation/08_belief_grasp_selection.py` | `python examples/manipulation/08_belief_grasp_selection.py` | pose belief -> grasp choice -> miss -> update -> retry |
 
 ## Embodied AI