pith. sign in

arxiv: 2606.24991 · v1 · pith:4JFHGVD3new · submitted 2026-06-23 · 📡 eess.SY · cs.LG· cs.SY· math.OC

Solving Markov Decision Processes with Future Information via MPC

Shambhuraj Sawant , Akhil S Anand , Dirk Reinhardt , Sebastien Gros This is my paper

Pith reviewed 2026-06-25 21:57 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SYmath.OC
keywords Model Predictive ControlMarkov Decision ProcessesReinforcement LearningFuture InformationOptimal PolicyFunction ApproximationControl Systems
0
0 comments X

The pith

A parameterized MPC exactly represents the optimal value functions and policy of an MDP with future information under specific structural requirements.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper shows how to incorporate future information such as forecasts or reference trajectories into model predictive control for Markov decision processes. It identifies the conditions needed for the MPC to exactly match the optimal policy and value function rather than approximate it. This matters for problems where future data is known at decision time, allowing the use of RL to tune the MPC parameters while keeping the benefits of constraint handling. The method is demonstrated on a point-mass racing task with future reference information.

Core claim

The paper establishes the structural requirements under which a parameterized MPC can exactly represent the optimal value functions and policy of an MDP with future information. Such a parameterized MPC can then serve as a structured function approximator whose parameters are learned using reinforcement learning.

What carries the argument

Parameterized model predictive control that includes future information in the state and optimization to match the augmented MDP structure.

If this is right

  • The parameters of the MPC can be adjusted via RL to achieve optimality for MDPs with future information.
  • This approach preserves the ability to enforce constraints and embed domain knowledge in the planning.
  • It applies to real-world sequential decision problems involving forecasts, prices, or trajectories.
  • The illustration on the racing task shows practical feasibility for such augmented MDPs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This structure might allow similar exact representations in other optimization-based controllers beyond MPC.
  • Extending to cases with uncertain future information could be a natural next step, though not addressed here.
  • Connections to other RL methods for handling partial observability or exogenous inputs may exist.

Load-bearing premise

The future information must be available perfectly at decision time, and the MDP must admit an exact MPC parameterization without approximation error.

What would settle it

Finding an MDP with future information where no choice of MPC parameters can exactly reproduce the optimal value function and policy would falsify the claim.

Figures

Figures reproduced from arXiv: 2606.24991 by Akhil S Anand, Dirk Reinhardt, Sebastien Gros, Shambhuraj Sawant.

Figure 1
Figure 1. Figure 1: Closed-loop performance and state trajectory for [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
read the original abstract

Model Predictive Control (MPC) is widely used in industrial and robotic systems for enforcing constraints and embedding domain knowledge through finite-horizon optimization-based planning. However, despite these strengths, an MPC scheme typically does not yield optimal policies for sequential decision-making problems formulated as Markov Decision Processes (MDPs). Recent combinations of MPC with Reinforcement Learning (RL) alleviate this issue by treating MPC as a parameterized model of the optimal policy of an MDP and adjusting its parameters using data. While these approaches typically consider classical MDPs, many real-world problems include future information--such as forecasts, prices, or reference trajectories--at decision time, which must be included in the MDP state for optimal decision-making. Current MPC-RL approaches do not directly account for this augmented-state structure, raising the question of how to incorporate future information into MPC to obtain an optimal policy. This work establishes the structural requirements under which a parameterized MPC can exactly represent the optimal value functions and policy of an MDP with future information. We further demonstrate that such a parameterized MPC can serve as a structured function approximator, with its parameters learned using RL. The approach is illustrated on a point-mass racing task with future reference information.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper claims that structural requirements exist under which a parameterized MPC exactly represents the optimal value functions and policy of an MDP whose state is augmented with future information (e.g., forecasts or references). It further asserts that the resulting MPC serves as a structured function approximator whose parameters can be learned via RL, and illustrates the approach on a point-mass racing task with future reference information.

Significance. If the claimed structural requirements can be derived and verified, the result would supply a concrete bridge between finite-horizon MPC and infinite-horizon optimality for MDPs that incorporate perfect future information at decision time. This would strengthen MPC-RL hybrids by guaranteeing exact representation rather than approximation, while retaining constraint-handling and domain-knowledge embedding. The racing-task illustration suggests immediate applicability to trajectory-tracking problems with reference previews.

major comments (2)
  1. [Abstract] Abstract: the central claim asserts existence of structural requirements allowing exact representation of optimal value functions and policy by parameterized MPC, yet supplies neither the requirements themselves, a derivation sketch, nor any equation relating the MPC cost, constraints, or horizon to the MDP Bellman operator. Without these elements the claim cannot be assessed.
  2. [Abstract] Abstract: the statement that the parameterized MPC 'can serve as a structured function approximator' with parameters 'learned using RL' is presented without specifying the RL algorithm, the loss, or how the future-information augmentation enters the MPC parameterization; this leaves the learning procedure undefined.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'future information--such as forecasts, prices, or reference trajectories' is introduced without a formal definition of how this information augments the MDP state or enters the MPC optimization.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the comments on the abstract. We address each point below, noting that the full derivations appear in the body of the manuscript while agreeing that the abstract can be clarified.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim asserts existence of structural requirements allowing exact representation of optimal value functions and policy by parameterized MPC, yet supplies neither the requirements themselves, a derivation sketch, nor any equation relating the MPC cost, constraints, or horizon to the MDP Bellman operator. Without these elements the claim cannot be assessed.

    Authors: The abstract summarizes the contribution at a high level. The structural requirements are formally derived in Section 3, a derivation sketch and proof appear with Theorem 1, and the explicit equations relating MPC cost, constraints, and horizon to the Bellman operator are given in Section 4. We will revise the abstract to include a concise statement of the key structural conditions. revision: yes

  2. Referee: [Abstract] Abstract: the statement that the parameterized MPC 'can serve as a structured function approximator' with parameters 'learned using RL' is presented without specifying the RL algorithm, the loss, or how the future-information augmentation enters the MPC parameterization; this leaves the learning procedure undefined.

    Authors: Section 5 specifies the policy-gradient RL algorithm, the expected cumulative reward as the loss, and the incorporation of future information as time-varying references within the MPC. We will revise the abstract to briefly indicate the RL procedure and parameterization. revision: yes

Circularity Check

0 steps flagged

No significant circularity; theoretical representation result is self-contained

full rationale

The paper's central claim is the derivation of structural requirements allowing a parameterized MPC to exactly represent optimal value functions and policies for MDPs with future information. This is framed as an existence/representation theorem rather than any fitted quantity, self-defined mapping, or prediction that reduces to its own inputs by construction. No equations, self-citations, or ansatzes are quoted that exhibit the enumerated circular patterns. The result is presented as independent of the specific fitted parameters later learned via RL, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of structural requirements allowing exact representation, but the abstract supplies no explicit free parameters, axioms, or invented entities; the domain assumption that future information must be included in the state is stated directly.

axioms (1)
  • domain assumption Future information such as forecasts or references is available at decision time and must be included in the MDP state for optimality.
    Explicitly stated in the abstract as the motivation for augmenting the MDP state.

pith-pipeline@v0.9.1-grok · 5752 in / 1173 out tokens · 21529 ms · 2026-06-25T21:57:49.025872+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

122 extracted references · 1 canonical work pages

  1. [1]

    2017 , publisher=

    Model predictive control: theory, computation, and design , author=. 2017 , publisher=

    2017

  2. [2]

    2018 , publisher=

    Reinforcement learning: An introduction , author=. 2018 , publisher=

    2018

  3. [3]

    IEEE Transactions on Automatic Control , volume=

    Data-driven economic NMPC using reinforcement learning , author=. IEEE Transactions on Automatic Control , volume=. 2019 , publisher=

    2019

  4. [4]

    IEEE Transactions on Automatic Control , year=

    Equivalence of Optimality Criteria for Markov Decision Process and Model Predictive Control , author=. IEEE Transactions on Automatic Control , year=

  5. [5]

    , journal=

    Hewing, Lukas and Kabzan, Juraj and Zeilinger, Melanie N. , journal=. Cautious Model Predictive Control Using Gaussian Process Regression , year=

  6. [6]

    Proceedings of the 2nd Conference on Learning for Dynamics and Control , pages =

    Bayesian model predictive control: Efficient model exploration and regret bounds using posterior sampling , author =. Proceedings of the 2nd Conference on Learning for Dynamics and Control , pages =. 2020 , volume =

    2020

  7. [7]

    Machine learning , volume=

    Q-learning , author=. Machine learning , volume=. 1992 , publisher=

    1992

  8. [8]

    Advances in neural information processing systems , volume=

    Policy gradient methods for reinforcement learning with function approximation , author=. Advances in neural information processing systems , volume=

  9. [9]

    International conference on machine learning , pages=

    Deterministic policy gradient algorithms , author=. International conference on machine learning , pages=. 2014 , organization=

    2014

  10. [10]

    Online Optimization of Large Scale Systems , pages=

    Sensitivity analysis and real-time optimization of parametric nonlinear programming problems , author=. Online Optimization of Large Scale Systems , pages=. 2001 , publisher=

    2001

  11. [11]

    2021 60th IEEE Conference on Decision and Control (CDC) , pages=

    MPC-based reinforcement learning for a simplified freight mission of autonomous surface vehicles , author=. 2021 60th IEEE Conference on Decision and Control (CDC) , pages=. 2021 , organization=

    2021

  12. [12]

    2023 European Control Conference (ECC) , pages=

    A Painless Deterministic Policy Gradient Method for Learning-based MPC , author=. 2023 European Control Conference (ECC) , pages=. 2023 , organization=

    2023

  13. [13]

    arXiv preprint arXiv:1701.07274 , year=

    Deep reinforcement learning: An overview , author=. arXiv preprint arXiv:1701.07274 , year=

    Pith/arXiv arXiv

  14. [14]

    The International Journal of Robotics Research , volume=

    Reinforcement learning in robotics: A survey , author=. The International Journal of Robotics Research , volume=. 2013 , publisher=

    2013

  15. [15]

    Advances in neural information processing systems , volume=

    Gradient descent for general reinforcement learning , author=. Advances in neural information processing systems , volume=

  16. [16]

    IEEE Open Journal of Control Systems , volume=

    Convex neural network-based cost modifications for learning model predictive control , author=. IEEE Open Journal of Control Systems , volume=. 2022 , publisher=

    2022

  17. [17]

    IEEE/ASME transactions on mechatronics , volume=

    Variable stiffness actuators: Review on design and components , author=. IEEE/ASME transactions on mechatronics , volume=. 2015 , publisher=

    2015

  18. [18]

    IEEE Robotics & Automation Magazine , volume=

    Compliant actuator designs , author=. IEEE Robotics & Automation Magazine , volume=

  19. [19]

    Hybrid position/force control of manipulators , author=. ASME, J. of Dynamic Systems, Measurement, and Control , volume=

  20. [20]

    Frontiers in Robotics and AI , volume=

    Variable Impedance Control and Learning—A Review , author=. Frontiers in Robotics and AI , volume=. 2020 , publisher=

    2020

  21. [21]

    IEEE Transactions on Control Systems Technology , volume=

    Force tracking impedance control of robot manipulators under unknown environment , author=. IEEE Transactions on Control Systems Technology , volume=. 2004 , publisher=

    2004

  22. [22]

    2015 IEEE international conference on robotics and automation (ICRA) , pages=

    Control of generalized contact motion and force in physical human-robot interaction , author=. 2015 IEEE international conference on robotics and automation (ICRA) , pages=. 2015 , organization=

    2015

  23. [23]

    IEEE Robotics and Automation Letters , volume=

    Learning variable impedance control for contact sensitive tasks , author=. IEEE Robotics and Automation Letters , volume=. 2020 , publisher=

    2020

  24. [24]

    2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Variable impedance control in end-effector space: An action space for reinforcement learning in contact-rich tasks , author=. 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2019 , organization=

    2019

  25. [25]

    Robotics: Science and Systems VI , volume=

    Variable impedance control a reinforcement learning approach , author=. Robotics: Science and Systems VI , volume=. 2011 , publisher=

    2011

  26. [26]

    Springer handbook of robotics , pages=

    Force control , author=. Springer handbook of robotics , pages=. 2016 , publisher=

    2016

  27. [27]

    2016 XIII Latin American Robotics Symposium and IV Brazilian Robotics Symposium (LARS/SBR) , pages=

    Understanding the implementation of impedance control in industrial robots , author=. 2016 XIII Latin American Robotics Symposium and IV Brazilian Robotics Symposium (LARS/SBR) , pages=. 2016 , organization=

    2016

  28. [28]

    The International Journal of Advanced Manufacturing Technology , volume=

    Compliant motion control of robots by using variable impedance , author=. The International Journal of Advanced Manufacturing Technology , volume=. 1992 , publisher=

    1992

  29. [29]

    Proceedings of the 28th International Conference on machine learning (ICML-11) , pages=

    PILCO: A model-based and data-efficient approach to policy search , author=. Proceedings of the 28th International Conference on machine learning (ICML-11) , pages=. 2011 , organization=

    2011

  30. [30]

    Data-Efficient Machine Learning workshop, ICML , volume=

    Improving PILCO with Bayesian neural network dynamics models , author=. Data-Efficient Machine Learning workshop, ICML , volume=

  31. [31]

    arXiv preprint arXiv:1606.01540 , year=

    Openai gym , author=. arXiv preprint arXiv:1606.01540 , year=

    Pith/arXiv arXiv

  32. [32]

    2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)(IEEE Cat

    Design and use paradigms for gazebo, an open-source multi-robot simulator , author=. 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)(IEEE Cat. No. 04CH37566) , volume=. 2004 , organization=

    2004

  33. [33]

    doi:10.5281/zenodo.4320612 , url =

    Saif Sidhik , title =. doi:10.5281/zenodo.4320612 , url =

    work page doi:10.5281/zenodo.4320612

  34. [34]

    IEEE Transactions on Systems, Man, and Cybernetics , volume=

    Compliance and force control for computer controlled manipulators , author=. IEEE Transactions on Systems, Man, and Cybernetics , volume=. 1981 , publisher=

    1981

  35. [35]

    1984 American control conference , pages=

    Impedance control: An approach to manipulation , author=. 1984 American control conference , pages=. 1984 , organization=

    1984

  36. [36]

    Proceedings of 1995 IEEE International Conference on Robotics and Automation , volume=

    Variable impedance control of a robot for cooperation with a human , author=. Proceedings of 1995 IEEE International Conference on Robotics and Automation , volume=. 1995 , organization=

    1995

  37. [37]

    Robotics and Autonomous Systems , volume=

    An introductory review of active compliant control , author=. Robotics and Autonomous Systems , volume=. 2019 , publisher=

    2019

  38. [38]

    , author=

    A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms. , author=. J. Mach. Learn. Res. , volume=

  39. [39]

    Journal of Intelligent & Robotic Systems , volume=

    Survey of model-based reinforcement learning: Applications on robotics , author=. Journal of Intelligent & Robotic Systems , volume=. 2017 , publisher=

    2017

  40. [40]

    arXiv preprint arXiv:1907.02057 , year=

    Benchmarking model-based reinforcement learning , author=. arXiv preprint arXiv:1907.02057 , year=

    Pith/arXiv arXiv 1907

  41. [41]

    IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) , volume=

    Impedance learning for robotic contact tasks using natural actor-critic algorithm , author=. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) , volume=. 2009 , publisher=

    2009

  42. [42]

    The International Journal of Robotics Research , volume=

    Learning variable impedance control , author=. The International Journal of Robotics Research , volume=. 2011 , publisher=

    2011

  43. [43]

    Autonomous Robots , volume=

    Learning motions from demonstrations and rewards with time-invariant dynamical systems based policies , author=. Autonomous Robots , volume=. 2018 , publisher=

    2018

  44. [44]

    2019 International Conference on Robotics and Automation (ICRA) , pages=

    Reinforcement learning on variable impedance controller for high-precision robotic assembly , author=. 2019 International Conference on Robotics and Automation (ICRA) , pages=. 2019 , organization=

    2019

  45. [45]

    Sensors , volume=

    Efficient force control learning system for industrial robots based on variable impedance control , author=. Sensors , volume=. 2018 , publisher=

    2018

  46. [46]

    Journal of Intelligent & Robotic Systems , volume=

    Model-based reinforcement learning variable impedance control for human-robot collaboration , author=. Journal of Intelligent & Robotic Systems , volume=. 2020 , publisher=

    2020

  47. [47]

    arXiv preprint arXiv:1805.12114 , year=

    Deep reinforcement learning in a handful of trials using probabilistic dynamics models , author=. arXiv preprint arXiv:1805.12114 , year=

    Pith/arXiv arXiv

  48. [48]

    2022 IEEE/SICE International Symposium on System Integration (SII) , pages=

    Evaluation of Variable Impedance-and Hybrid Force/MotionControllers for Learning Force Tracking Skills , author=. 2022 IEEE/SICE International Symposium on System Integration (SII) , pages=. 2022 , organization=

    2022

  49. [49]

    Applied Sciences , volume=

    Variable compliance control for robotic peg-in-hole assembly: A deep-reinforcement-learning approach , author=. Applied Sciences , volume=. 2020 , publisher=

    2020

  50. [50]

    IEEE Robotics and Automation Letters , volume=

    Learning force control for contact-rich manipulation tasks with rigid position-controlled robots , author=. IEEE Robotics and Automation Letters , volume=. 2020 , publisher=

    2020

  51. [51]

    Conference on Robot Learning , pages=

    SCAPE: Learning Stiffness Control from Augmented Position Control Experiences , author=. Conference on Robot Learning , pages=. 2022 , organization=

    2022

  52. [52]

    Handbook of statistics , volume=

    The cross-entropy method for optimization , author=. Handbook of statistics , volume=. 2013 , publisher=

    2013

  53. [53]

    Advances in Neural Information Processing Systems , volume=

    Constrained cross-entropy method for safe reinforcement learning , author=. Advances in Neural Information Processing Systems , volume=

  54. [54]

    2012 IEEE/RSJ international conference on intelligent robots and systems , pages=

    Mujoco: A physics engine for model-based control , author=. 2012 IEEE/RSJ international conference on intelligent robots and systems , pages=. 2012 , organization=

    2012

  55. [55]

    Journal of Neurophysiology , volume=

    Stiffness as a control factor for object manipulation , author=. Journal of Neurophysiology , volume=. 2019 , publisher=

    2019

  56. [56]

    Proceedings of the Royal Society of London

    The series elastic component of muscle , author=. Proceedings of the Royal Society of London. Series B, Biological Sciences , pages=. 1950 , publisher=

    1950

  57. [57]

    Journal of Neuroscience , volume=

    Posture control and trajectory formation during arm movement , author=. Journal of Neuroscience , volume=. 1984 , publisher=

    1984

  58. [58]

    Journal of neuroscience , volume=

    An organizing principle for a class of voluntary movements , author=. Journal of neuroscience , volume=. 1984 , publisher=

    1984

  59. [59]

    Proceedings 2002 IEEE International Conference on Robotics and Automation (Cat

    Cartesian impedance control techniques for torque controlled light-weight robots , author=. Proceedings 2002 IEEE International Conference on Robotics and Automation (Cat. No. 02CH37292) , volume=. 2002 , organization=

    2002

  60. [60]

    Proceedings

    Optimal variable impedance control for a robot and its application to lifting an object with a human , author=. Proceedings. 11th IEEE International Workshop on Robot and Human Interactive Communication , pages=. 2002 , organization=

    2002

  61. [61]

    Robotics and Autonomous Systems , volume=

    Force-based variable impedance learning for robotic manipulation , author=. Robotics and Autonomous Systems , volume=. 2018 , publisher=

    2018

  62. [62]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Learning collaborative impedance-based robot behaviors , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  63. [63]

    2013 , publisher=

    Model predictive control , author=. 2013 , publisher=

    2013

  64. [64]

    2020 IEEE international conference on robotics and automation (ICRA) , pages=

    Model predictive impedance control , author=. 2020 IEEE international conference on robotics and automation (ICRA) , pages=. 2020 , organization=

    2020

  65. [65]

    Proceedings of 17th International Conference of the Engineering in Medicine and Biology Society , volume=

    Model predictive impedance control: application to human walking model , author=. Proceedings of 17th International Conference of the Engineering in Medicine and Biology Society , volume=. 1995 , organization=

    1995

  66. [66]

    Journal of motor behavior , volume=

    Model predictive impedance control: A model for joint movement , author=. Journal of motor behavior , volume=. 1997 , publisher=

    1997

  67. [67]

    Journal of mathematics and mechanics , pages=

    A Markovian decision process , author=. Journal of mathematics and mechanics , pages=. 1957 , publisher=

    1957

  68. [68]

    arXiv preprint arXiv:2006.16712 , year=

    Model-based reinforcement learning: A survey , author=. arXiv preprint arXiv:2006.16712 , year=

    arXiv 2006

  69. [69]

    International conference on machine learning , pages=

    Model-based active exploration , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  70. [70]

    International Conference on Machine Learning , pages=

    Planning to explore via self-supervised world models , author=. International Conference on Machine Learning , pages=. 2020 , organization=

    2020

  71. [71]

    International conference on machine learning , pages=

    Self-supervised exploration via disagreement , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  72. [72]

    arXiv preprint arXiv:2012.05909 , year=

    Blending mpc & value function approximation for efficient reinforcement learning , author=. arXiv preprint arXiv:2012.05909 , year=

    arXiv 2012

  73. [73]

    Proceedings of SAI Intelligent Systems Conference , pages=

    Deep reinforcement learning: an overview , author=. Proceedings of SAI Intelligent Systems Conference , pages=. 2016 , organization=

    2016

  74. [74]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Deep reinforcement learning that matters , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  75. [75]

    arXiv preprint arXiv:1802.10592 , year=

    Model-ensemble trust-region policy optimization , author=. arXiv preprint arXiv:1802.10592 , year=

    Pith/arXiv arXiv

  76. [76]

    Advances in Neural Information Processing Systems , volume=

    When to trust your model: Model-based policy optimization , author=. Advances in Neural Information Processing Systems , volume=

  77. [77]

    2021 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Sample Efficient Reinforcement Learning via Model-Ensemble Exploration and Exploitation , author=. 2021 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2021 , organization=

    2021

  78. [78]

    International Conference on Machine Learning , pages=

    Pc-mlp: Model-based reinforcement learning with policy cover guided exploration , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  79. [79]

    International Conference on Machine Learning , pages=

    Model-based Reinforcement Learning for Continuous Control with Posterior Sampling , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  80. [80]

    Proceedings of international conference on robotics and automation , volume=

    A comparison of direct and model-based reinforcement learning , author=. Proceedings of international conference on robotics and automation , volume=. 1997 , organization=

    1997

Showing first 80 references.