{"id":"5ee5c08d-9692-4d54-8ce1-14292ad54b1b","arxiv_id":"2504.17901","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"TASP integrates composable interaction primitives into the ATAM task and motion planner so that pre-trained, force-controlled, and black-box robot skills can be chained with motion plans to solve long-horizon tasks.","lead":"This paper introduces TASP, a hierarchical planner that lets a robot combine ordinary path planning with pre-existing skills, including force-controlled and black-box actions, by wrapping each skill with planned approach and exit motions. Two real robots use it to finish long, multi-step jobs like spreading peanut butter and erasing a whiteboard across rooms.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"TASP's low-level refinement verifies only the kinematic envelope λ(I_a), not the full initiation condition I_a; because λ drops observation predicates, the planner can commit to a skill whose true preconditions are false, so the claimed validity of plans for new scenes is not established.","rationale":"In good faith, the paper's contribution is real: it packages non-kinematic skills inside CIPs and uses ATAM-style backtracking to sample head and tail endpoints for long-horizon composition. The two real-robot demonstrations, including the non-monotonic mobile manipulation task, show that the idea works when the skills' observation conditions happen to hold. The load-bearing weakness is the gap between property (iii) and the implemented refinement. Property (iii) in Sec. III explicitly requires selecting a state satisfying the full initiation condition I_a, but the CIP abstraction λ in Sec. IV-A deliberately removes observation functions, and Sec. IV-B only checks λ(I_a) during refinement. This is not an external disagreement with consensus; it is an internal weakening of the paper's own criterion. The paper states that the sampled head end must allow successful skill execution 'given the other objects in the environment,' but provides no algorithmic procedure, no verification step, and no recovery mechanism for that condition. The single-run experiments cannot disambiguate whether the success came from the planner's guarantees or from a carefully staged scene in which the abstracted observations were consistent. This concern does not warrant rejection because the paper is framed as a demonstration and does not claim formal guarantees; however, it does mean the strongest wording ('valid long-horizon plans for new scenes') overstates what is shown. The reader's conditional verdict already captures this, so I recommend no change: keep CONDITIONAL, with the requested revisions focusing on either a stated verification of full preconditions, a bounded execution monitor with replanning, or a rephrased claim that does not imply validity beyond the demonstrated scenarios.","tokens_in":10750,"tokens_out":7476,"duration_ms":89661,"concrete_test":"Implement TASP's refinement loop on a minimal hybrid domain with one skill Wipe(?surface) whose initiation condition I_a includes surface_wet(?surface), but whose CIP envelope λ(I_a) contains only the surface pose and the robot configuration. Set the initial symbolic state to surface_wet=false and surface_clean=false, goal surface_clean=true, so that Wipe is the only way to reach the goal; give the Wipe policy a failure condition when the surface is dry. If TASP returns a Wipe plan and the policy fails during execution, the λ-only refinement is unsound. Repeat for 10 randomized surface poses to confirm the failure is not a perception artifact.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's third composability property (Sec. III) requires selecting x_i |= I_a and x_j |= β_a such that the skill policy π_a can be successfully executed from x_i to x_j. However, Sec. IV-B replaces this with the weaker condition that the head motion plan reaches x_n |= λ(I_a), where λ is defined in Sec. IV-A as the projection that abstracts object-specific observation functions out of the initiation condition. The text says the head end must be 'such that π_ai can be successfully executed given the other objects in the environment,' but no mechanism is described that verifies this: ATAM's pose generators sample configurations in the kinematic envelope, and any observation predicates in the symbolic state are taken from the provided initial model rather than checked at the sampled configuration. Consequently, a skill whose success depends on Φ_o (e.g., OpenDoor requiring door_unlocked, Erase requiring board_dirty, or OpenDrawer requiring drawer_closed) can be planned even when the observation condition is false in the real scene, and the paper documents no recovery or replanning path. The real-world experiments are single runs with hand-initialized symbolic states and static kinematic models for articulated objects (Sec. V-A.2), so they do not exercise this gap. Without a stated way to ensure x_n ∈ I_a, or to detect and recover from skill failure, the central claim that TASP 'yields valid long-horizon plans for new scenes' is not fully supported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes Task and Skill Planning (TASP), a hierarchical planning approach that integrates pre-existing, heterogeneous robot skills—including learned, force-controlled, and black-box policies—into a task-and-motion-planning (TAMP) framework. The key mechanism is to package each skill as a Composable Interaction Primitive (CIP), which adds a kinematic head motion plan into the skill's initiation set and a tail motion plan out of its termination set. The paper introduces a projection λ that abstracts object-specific observation predicates out of the initiation and termination conditions, yielding 'kinematic envelopes' that motion planning can target. TASP then uses the ATAM TAMP solver, modified to call skill policies instead of motion planners during refinement, to compose CIPs into long-horizon plans. The authors validate the approach in real-world experiments on a bimanual manipulator (opening a jar, scooping and spreading peanut butter) and on a Boston Dynamics Spot mobile manipulator (multi-room task involving opening a door, opening a drawer, picking and placing an eraser, closing the door, and erasing a whiteboard).","tokens_in":11009,"tokens_out":4077,"duration_ms":42633,"significance":"If the central claim holds, TASP is a meaningful step beyond classical TAMP: it allows non-kinematic skills, including sustained-contact and black-box policies, to be reused in long-horizon plans without retraining. The formalization of a hybrid robot planning problem (Def. 3) and the explicit identification of the three composability properties for skills (Sec. III) are useful conceptual contributions. The real-robot demonstrations are genuinely impressive in scope and show that the composed skills execute end-to-end on different platforms. However, the significance is tempered by a gap between the kinematic envelope λ(I_a) and the full initiation condition I_a, which the manuscript does not close; as a result, the claim of producing 'valid long-horizon plans for new scenes' is not fully supported. The paper does not ship machine-checked proofs or code, so the empirical demonstration carries much of the burden, and the experimental evidence is presently anecdotal (single runs, no quantitative metrics).","major_comments":[{"comment":"The composition mechanism verifies only the kinematic envelope λ(I_a), not the full initiation condition I_a. In Sec. IV-A, λ is defined as the projection that removes object-specific observation functions Φ_o, so a head motion plan ending at a configuration x_n with x_n |= λ(I_a) says nothing about whether observation predicates such as 'board is dirty' or 'drawer is closed' hold at that configuration. The text in Sec. IV-B states that x_n must be 'such that π_ai can be successfully executed given the other objects in the environment,' but no mechanism is described to verify this condition: ATAM's pose generators sample configurations in the kinematic envelope, and observation predicates are taken from the provided initial symbolic state rather than checked at the sampled configuration. Consequently, the planner can commit to a skill whose true preconditions are false (e.g., OpenDoor on a locked door, Erase on a clean whiteboard), and the paper documents no detection or recovery path. Sec. V-A.2 acknowledges indirectly that articulated objects are modeled only before and after skills, not during them, and the experiments are initialized with hand-set symbolic states, so the gap is not exercised. Because the central claim is that TASP 'yields valid long-horizon plans for new scenes' (Abstract and Sec. VI), this is a load-bearing issue. The authors should either add a mechanism to verify I_a at the sampled configuration, add a replanning/recovery loop when a skill fails, or substantially weaken the validity claim to a demonstration given correct symbolic state.","section":"Sec. IV-A/IV-B and Def. 4"},{"comment":"The TASP algorithm itself is not fully specified, which hampers reproducibility and makes it hard to separate the contribution from simply hand-scheduling the demonstrated skills. The text says ATAM is 'modified' to use pre-trained skill policies and that entity abstraction creates discrete parameters for head/tail motion plans and the skill policy, but it does not describe how CIP endpoints are sampled by ATAM's pose generators, how the condition 'π_ai can be successfully executed given the other objects in the environment' is checked during refinement, or how backtracking interacts with skill execution versus pure kinematic motion planning. No pseudocode, algorithm box, or formal specification of the modified Γ (inverse abstraction) is provided. Given that the paper's main technical claim is that ATAM efficiently composes CIPs, this underspecification is a major weakness. I recommend adding a concrete algorithm description, including the state space of the symbolic planner, the sampling distribution for CIP endpoints, and the failure-triggering conditions for backtracking.","section":"Sec. IV-B"},{"comment":"The experimental validation is demonstrative rather than evidential. Each task is performed in a single run, with no quantitative success metrics, no repeated trials, no failure cases, and no comparison to a baseline planner. The multi-room mobile manipulation experiment exhibits non-monotonic structure, which is a nice feature, but the central claim about 'valid plans for new scenes' requires at least some evidence that the planner does not rely on fragile initialization or on observation predicates being true by construction. The paper should report, at minimum, the number of runs and their outcomes, and should describe what happens when a skill's initiation predicate is false (e.g., the whiteboard is already clean or the drawer is already open). Without such data, the real-robot results support the weaker claim that the composed skills can execute in favorably initialized scenes, not the stronger claim of general validity.","section":"Sec. V"}],"minor_comments":[{"comment":"The notation in Definition 4 is asymmetric: the head motion plan h goes from x not satisfying I_a to x satisfying λ(I_a), while the tail motion plan t goes from x satisfying β_a to x not satisfying λ(β_a). Please clarify whether the source conditions for h and t also use the full predicates or their kinematic envelopes, and make the notation consistent.","section":"Sec. IV-A, Def. 4"},{"comment":"Definition 3 defines a solution as a sequence of skills that reaches x_n ∈ X_g, but the relation between X_g and the PDDL goal conditions G, used in the experiments, is never formally spelled out. A brief sentence connecting Def. 3 to the experimental problem setting would help.","section":"Sec. III, Def. 3"},{"comment":"There are several typographical and formatting errors, including 'ComposableInteraction Primitives' missing a space, inconsistent hyphenation of 'task and skill planning', and footnote 3 indicating omitted GoTo instances that are then still described in the text. A careful proofread is recommended.","section":"Abstract and throughout"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing to know: this is an engineering-integration paper, not a new theory paper, and it is mostly honest about that despite a few overreaching sentences. The genuinely new part is coupling Composable Interaction Primitives (CIPs) with the ATAM backtracking TAMP search so that pre-existing force-controlled, playback, and black-box skills can be chained with motion planning. That is a real and useful capability for generalist robots. The two real-robot demonstrations—bimanual jar-opening/spreading and the Spot multi-room erase task—are also legitimately impressive as system demos, especially the non-monotonic door/drawer structure.\n\nWhat the paper does well: it gives a clean formalization of the hybrid robot planning problem, it correctly identifies the three composability properties a skill wrapper needs, and it does not oversell the individual components—CIPs and ATAM are both attributed to prior work. The skill inventory is genuinely heterogeneous (motion planning, impedance control, force control, trajectory playback, a true black-box Spot SDK capability), which makes the integration claim concrete.\n\nNow the soft spots, in proportion. The stress-test concern about λ(I_a) is real and it lands. The paper's own Property (iii) requires selecting x_i |= I_a, but the implementation only verifies x_n |= λ(I_a), the projection that drops observation predicates like board_dirty or door_unlocked. Section IV-B says the head end must be 'such that π_ai can be successfully executed given the other objects,' but no mechanism is described that checks this. The real experiments use hand-initialized symbolic states and static kinematic models of articulated objects before/after the skill, so they never exercise the gap. That means the central claim about 'valid long-horizon plans for new scenes' is not fully supported as stated; the paper should either restrict the claim to scenes where the observation predicates are known to hold, or add a detection/replanning loop for skill failure.\n\nThe other issues are more conventional. The TASP algorithm itself is underspecified: how ATAM's pose generators are constrained to sample head/tail endpoints that satisfy the full skill preconditions, and how backtracking interacts with skill policy failures, is described only at a high level. There is no quantitative evaluation, no baseline comparison, and no released code or data, so the demos are single-run anecdotes. The 'first approach' claim in Section VI is also stronger than the evidence—several prior papers have combined learned skills with TAMP, and the discussion of that literature is brief. Self-citation is not a problem here; the CIP and ATAM references are appropriate and the integration is externally demonstrated.\n\nBottom line: the paper deserves a serious referee. The integration is plausible, the demos are informative, and the formal gap is fixable with an honest assumption statement or a failure-handling mechanism. I would send it to review with a request for a full algorithm specification and a toned-down validity claim. My own verdict would be conditional acceptance, not rejection.","headline":"TASP is a genuinely useful integration of CIPs with the ATAM TAMP loop, with real-robot demos that show the pieces working end-to-end; the main weakness is that the formal 'validity' claim only checks the kinematic envelope λ(I_a), not the full initiation condition, and the paper offers no recovery when the dropped observation predicates are false.","tokens_in":11582,"tokens_out":1574,"would_cite":true,"duration_ms":18414,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that wrapping pre-existing robot skills—learned, force-controlled, and black-box—with head and tail motion plans lets a hierarchical task-and-motion planner compose them into long-horizon plans without retraining.","keywords":["task and motion planning","composable interaction primitives","hierarchical robot planning","hybrid robot planning","general-purpose skills","force control","black-box skills","long-horizon manipulation"],"falsifier":"Run the planner on a scene where the perceptual precondition is false while the kinematic envelope is satisfied—for instance, a whiteboard that is already clean fed to an erase skill—and observe whether the planner still emits a plan whose skill step fails without any recovery branch; alternatively, place a small obstacle inside the envelope that the head plan avoids but that the skill policy collides with, and see whether the skill fails at execution time.","tokens_in":10497,"feed_emoji":"🤖","tokens_out":5228,"duration_ms":50803,"temperature":0.7,"pith_summary":"The paper introduces Task and Skill Planning (TASP), a hierarchical planning method that lets long-horizon robot tasks be solved by mixing ordinary motion planning with pre-existing skills that are not kinematic: learned policies, force-controlled routines, trajectory playback, and black-box robot commands. The central move is to wrap each such skill in a Composable Interaction Primitive (CIP), consisting of a head motion plan that drives the robot to a position-only version of the skill's start condition, the skill itself, and a tail motion plan that returns the robot to free space. The wrapped skills are handed to a task-and-motion planner that refines symbolic actions, preserving the object-centric failure reasoning of classical TAMP. Real-robot demonstrations on a bimanual manipulator and a Boston Dynamics Spot show the method producing valid long-horizon plans, including a multi-room task with a non-monotonic door sequence. If correct, this is the first hierarchical planner beyond classical TAMP that composes kinematic and non-kinematic general-purpose skills.","feed_headline":"Planner wraps any robot skill in motion plans to solve long tasks","feed_subtitle":"Head and tail trajectories around force-controlled and learned skills let one planner reuse them without retraining.","key_machinery":"The load-bearing object is the Composable Interaction Primitive (CIP): a skill defined by its parameters, initiation condition, termination condition, and policy, together with a head motion plan from any configuration not in the initiation set to a configuration satisfying the skill's kinematic envelope, and a tail motion plan from a termination configuration to a configuration outside that envelope. The kinematic envelope is the initiation or termination condition projected to spatial functions only, dropping observation predicates such as 'board is dirty' or 'drawer is open'. This turns a non-kinematic skill into something a task-and-motion planner can refine, because the planner only needs to sample the envelope's end configurations, and the head and tail plans are ordinary collision-free trajectories. Composition is carried out by ATAM, whose entity abstraction creates discrete symbols for CIP endpoints and whose backtracking search resamples endpoints when the symbolic plan cannot be refined.","core_discovery":"The paper's central claim is that the three properties needed to compose general-purpose skills—reach the skill's initiation set, execute the skill safely, and move back to free space—can be realized by CIPs, and that an existing task-and-motion planning algorithm (ATAM) can be used to search over the continuous endpoints of CIP head and tail plans. Once a skill is abstracted to a kinematic envelope, the planner treats the head and tail motion plans as refinements of a symbolic action and uses backtracking to resample end configurations when refinement fails. The paper demonstrates this by solving two real planning problems with heterogeneous skill inventories, one on a bimanual manipulator and one on a mobile manipulator, and states that this is the first approach beyond classical TAMP to perform hierarchical planning with kinematic and non-kinematic general-purpose robot skills.","pith_inferences":["Because CIPs only require a kinematic envelope and motion plans, the same wrapping likely applies to skills with stochastic learned policies, provided the envelope condition is met; this is not tested in the paper.","A natural next step, which the paper leaves implicit, is to learn the initiation and termination sets themselves from data so the symbolic model does not have to be hand-authored.","A testable extension would measure how planning success degrades as the gap grows between the kinematic envelope and the true conditions under which the skill policy actually succeeds, for example by adding clutter near the skill's interaction zone.","The current method is deterministic; extending it to stochastic skills would require belief-state reasoning, likely going beyond the ATAM refinement loop used here."],"forward_implications":["Pre-trained skills no longer need retraining for new scenes; the planner supplies collision-free access and exit motion around them.","TAMP's object-centric reasoning is preserved, so skills can be sequenced by their symbolic effects while low-level feasibility is checked by sampling.","Non-kinematic skills, including force-controlled contact, trajectory playback, and black-box robot capabilities, become composable modules in long-horizon plans.","Solvable tasks extend to multi-room, non-monotonic problems where the same physical object is manipulated through several different skill types.","If the key assumption holds, the approach provides a blueprint for integrating existing skill libraries into general-purpose robot planning stacks."],"supporting_citations":[{"why":"Defines Composable Interaction Primitives (CIPs), the policy wrapper with head and tail motion plans that TASP adopts.","marker":"[1]"},{"why":"Supplies the ATAM task-and-motion planning algorithm, entity abstraction, and backtracking-based refinement that TASP modifies.","marker":"[24]"},{"why":"Surveys integrated task and motion planning and frames the classical assumption that TAMP actions reduce to kinematic motion planning, which TASP targets.","marker":"[7]"},{"why":"Provides the CBiRRT constraint-aware planner used to implement the motion-planning-based Pick skill.","marker":"[3]"},{"why":"Supplies the Open Motion Planning Library planners used by MoveIt! for Pick, Place, and GoTo skills.","marker":"[27]"}],"fun_headline_variants":["Head-tail motion plans bridge black-box robot skills","Planner composes any skill with head and tail plans","Stitching skills: head-tail plans for long tasks","Reuse any robot skill with head-tail motion plans","Hierarchical planner wraps skills in motion plans"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes that once the robot reaches a configuration in the skill's start region, computed from positions only and ignoring whether the board is actually dirty or the drawer is actually open, the learned or black-box skill policy will execute successfully.","fun_headline_variants_meta":{"raw":{"variants":["Head-tail motion plans bridge black-box robot skills","Planner composes any skill with head and tail plans","Stitching skills: head-tail plans for long tasks","Reuse any robot skill with head-tail motion plans","Hierarchical planner wraps skills in motion plans"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000387,"raw_usage":{"total_tokens":2012,"prompt_tokens":884,"completion_tokens":1128,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":500,"completion_tokens_details":{"reasoning_tokens":1054}},"tokens_in":500,"tokens_out":1128,"duration_ms":8259,"temperature":1.0,"reasoning_tokens":1054,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:29:58.868027+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the planner on a scene where the perceptual precondition is false while the kinematic envelope is satisfied—for instance, a whiteboard that is already clean fed to an erase skill—and observe whether the planner still emits a plan whose skill step fails without any recovery branch; alternatively, place a small obstacle inside the envelope that the head plan avoids but that the skill policy collides with, and see whether the skill fails at execution time.","supporting_citations":[{"cited_title":"Composable interac- tion primitives: A structured policy class for efficiently learning sustained-contact manipulation skills,","cited_arxiv_id":null,"evidence_quote":"Defines Composable Interaction Primitives (CIPs), the policy wrapper with head and tail motion plans that TASP adopts."},{"cited_title":"Anytime integrated task and motion policies for stochastic environments,","cited_arxiv_id":null,"evidence_quote":"Supplies the ATAM task-and-motion planning algorithm, entity abstraction, and backtracking-based refinement that TASP modifies."},{"cited_title":"Manipulation planning on constraint mani- folds,","cited_arxiv_id":null,"evidence_quote":"Provides the CBiRRT constraint-aware planner used to implement the motion-planning-based Pick skill."},{"cited_title":"The Open Motion Planning Library,","cited_arxiv_id":null,"evidence_quote":"Supplies the Open Motion Planning Library planners used by MoveIt! for Pick, Place, and GoTo skills."}],"review_version":1}