A Tool Registry

ATool Registry

Table 4 lists the Astribot S1 tool registry. Its entries follow the protocol of §3.3: physical capabilities use deployment backends, while lightweight context and interaction tools run in-process. Slash-separated lists are the admissible values; quoted strings are free-form descriptions; refs like desk_24 name Scene Graph entities.

Table 4. Tools deployed on the Astribot S1, grouped by function. evaluate_run is registered alongside them but invoked by the harness through a post-hook, never selected by the model (§4.2).

Tool	Description	Backend	Parameter	Value
Navigation
`navigate_to`	Approach an object or location.	SysNav [28]	`target`	e.g. “drink area”/desk_24
`move_base`	Translate or rotate within safety bounds.	Astribot S1 SDK	`direction`	forward/backward/left/right/turn_left/turn_right
			`distance_m`	0.03–0.80 m
			`angle_deg`	3–30°
Manipulation
`pick_up`	Grasp a localized object.	CaP-X/ACT/`π`_0.5 [32, 39, 10]	`obj`	e.g. “orange juice”/bottle_2
`place`	Place the held object on a table or in a box.	CaP-X/ACT/`π`_0.5	`target`	e.g. “table”/table_2
`open_drawer`	Open a drawer.	CaP-X/ACT/`π`_0.5	`container_ref`	e.g. cabinet_87
`open_drawer`	Open a drawer.	CaP-X/ACT/`π`_0.5	`drawer_level`	higher/lower
`close_drawer`	Close a drawer.	CaP-X/ACT/`π`_0.5	`container_ref`	e.g. cabinet_87
`close_drawer`	Close a drawer.	CaP-X/ACT/`π`_0.5	`drawer_level`	higher/lower
`trash_drop`	Drop the held object into a bin.	CaP-X/ACT/`π`_0.5	`trash_ref`	e.g. trash_can_54
Perception
`tilt_head`	Tilt the head camera to a given pitch angle.	Astribot S1 SDK	`pitch_deg`	0–60°
`get_object_relations`	Query an object's spatial and semantic relations.	Scene Graph	`obj`	e.g. cup_2
`get_object_relations`	Query an object's spatial and semantic relations.	Scene Graph	`relation` (optional)	on/inside/holding/near
`get_image`	Retrieve stored images of an object.	Scene Graph	`obj`	e.g. cup_2
Skills
`load_skill`	Load the instructions of a registered skill.	Thea	`name`	e.g. tidy-workspace
User Interaction
`query_user`	Request and await clarification.	Lark	`question`	e.g. “No plain water; which drink should I bring?”
			`candidate_refs` (optional)	e.g. bottle_3
			`observation_views` (optional)	e.g. torso_rgbd
`notify_user`	Send a non-blocking update.	Lark	`message`	e.g. “I found an empty orange-juice bottle on desk…”
`notify_user`	Send a non-blocking update.	Lark	`notification_type` (optional)	progress/warning/completion

[28]: Zhu et al. (2026), SysNav: Multi-Level Systematic Cooperation Enables Real-World, Cross-Embodiment Object Navigation. [32]: Fu et al. (2026), CaP-X: A framework for benchmarking and improving coding agents for robot manipulation. [39]: Zhao et al. (2023), Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. [10]: Black et al. (2025), π_0.5: A Vision-Language-Action Model with Open-World Generalization.